This PDF file contains the front matter associated with SPIE Proceedings Volume 12800, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

In order to promote the coupon distribution in the Web 3.0 scenario and explore new areas of marketing strategy, this research focuses on the user-delivery mechanism of coupons in different driving mobile scenes, introduces the data mining method, and proposes the fusion algorithm integrating Logistic Regression, SVM, Random Forest and Adaboost based on stacking model to build the prediction model of in-vehicle coupon acceptance behavior. By comparing the performance of the fusion model with four single models, the results show that the fusion framework can effectively improve the model training effect. In addition, this paper uses the statistical grouping method to analyze the driving distance and the types of coupons and puts forward the strategy of coupon delivery: coupons are issued to users within 15 minutes' drive of the shop, and users with a drive above 25 minutes will issue takeaway coupons instead.

The current multi-source heterogeneous sensory data fusion lacks the application of user feedback, resulting in low user satisfaction of the data fusion results. In this regard, a multi-source heterogeneous perceptual data fusion method for user feedback is proposed. A multi-source heterogeneous perceptual data storage model is built to complete the acquisition and storage of data. Uniform semantic annotation of the stored data. Obtain user relevance feedback. Based on the feedback information, the dynamic feedback fusion of multi-source heterogeneous perceptual data is completed. The experiments show that the average user satisfaction of the method is 0.9534, which is a significant improvement and has high practical application value.

Since wind turbine blades are exposed to harsh environment for a long time, they are very easy to produce defects and can cause serious economic losses, so this paper proposes a wind turbine blade defect detection method. Firstly, the background separation is performed on the acquired image, and the multi-scale Retinex algorithm is used to enhance the image. Then, findcontour and contourarea operators are used to extract the contour of the blade defect area. The experimental results are good and can meet the practical application requirements of burner blade defect detection.

In order to improve the properties of special coatings, many additives should be added in the preparation process. The properties of coating are affected by multi-dimensional parameters such as the content ratio of various additives and the type of resin material as the base of coating. In order to solve the problem of multi - additive special coating preparation dependent on experience, the preparation coating does not match the base material, the lack of effective guidance of mathematical models. The introduction of machine learning technology, through the analysis of historical preparation data. K-nearest neighbor algorithm (KNN) was used to establish the mathematical model of the composite interaction between the content of various additives and the contact angles of different resin materials. It is proved that this model has good prediction accuracy by using the actual data.

It is very dangerous to put the strip containing defects into use, and such defects are often highly susceptible to be misdiagnosed and missed in the actual inspection process due to the small difference from the background and large scale variation. Therefore, an improved surface defect detection algorithm based on YOLOv5s is proposed. Firstly, combined the idea of DenseNet network, a new C3 structure for the backbone network of the model is obtained by adding the convolution module inside the residual component and making multi-level dense connection to enhance feature reuse. Secondly, in the neck network, the original feature pyramid module is replaced by Generalized-FPN (GFPN) to achieve the full feature fusion. Finally, the experiments on NEU-DET show that the mean average precision (mAP) of the improved algorithm is 79.8%, and meets the needs of practical industrial detection.

Raven-II is a multifunctional surgical robot mainly used for minimally invasive surgery. This article aims to provide innovative ideas for Raven-II research through simulation verification and finite element analysis. Firstly, this article conducts SolidWorks 3D modeling of Raven-II and completes the model reproduction of the robot structure. Through the research and simple exposition of Raven II's kinematics analysis, a new method for solving the forward kinematics analysis is used in Python, which provides a new idea for robot research. The dynamics simulation of Raven-II was conducted in ADAMS simulation software, and its workspace and working mode were determined to meet the requirements of minimally invasive surgery, verifying the correctness of Raven-II structural design. Finally, finite element analysis and modal analysis were conducted on key components of the robot using ANSYS.

In order to solve the problem of multiple air refueling mission planning in the execution of long-range raiding missions, a multi-point air refueling model was established considering the timeliness and economy of the task. The differential evolution algorithm is improved by using the strategy of adaptively changing parameters, which makes it have faster search speed and computational efficiency. Simulation results show that the proposed method can effectively solve the problem of selecting fuel point in aircraft multi-point aerial refueling mission.

The transfer station of railway freight system, as an important department of freight undertaking and distribution, is responsible for the demand of freight transfer. Due to the influence of many factors such as the quantity and batch of transportation, not only the railway system needs to be planned and arranged, but also the highway logistics system needs to be arranged in order to improve the efficiency of logistics transportation, thus putting forward higher requirements for the location of transfer stations. To improve the competitiveness of the freight market and rationally plan the location of railway transfer stations, this paper constructs a neural network model based on multiple indicators of cost and path to optimize the location of railway transfer stations. The traditional ant colony algorithm is improved to solve the location of transfer stations, which is to achieve the purpose of efficient freight transfer and distribution. Through the analysis of simulation experiments, the improved ant colony algorithm can be adopted to solve the location path optimization of railway freight transfer index. Compared with the BP neural network and the traditional ant colony algorithm, it is verified that the method in this paper has better calculation accuracy and convergence effect. It proves that this method has the best overall integrity. It meets the efficiency of freight distribution, achieves the goal of the lowest cost, and provides technical verification results for the location of railway freight transfer stations and freight layout planning, which has good theoretical value.

With the flourishing development of the Internet of Things (IoT), IoT devices are becoming increasingly connected with people. IoT devices are constantly collecting and transmitting large amounts of private information, and there is a need for data access control and secure data sharing. Since Ciphertext Policy Attribute-Based Encryption (CP-ABE) is an end-to-end encryption solution, CP-ABE-based access control methods can effectively meet the data security sharing requirements in the IoT environment. However, existing CP-ABE access control methods are difficult to implement data access control. There are no standardized data access control solutions, and redundant storage of data can occur. This study addresses the issue that existing CP-ABE access control methods have difficulty in achieving data permission management. A CP-ABE access control scheme that supports data permission management is designed. To support data permission management, the data is partitioned into key-value pairs. A data permission management policy is designed based on the access control policy, which not only sets the access policy but also specifies the data unit scope of authorized users. The CP-ABE algorithm is used to encrypt the set of encrypted data unit keys in the policy, replacing the redundant storage of data units with keys. This study formulates a standardized data permission management scheme based on the CP-ABE access control scheme, effectively reducing data redundancy storages.

Wind tunnel experiment plays a very important role in the development of aircraft, but the existence of wind tunnel wall has a significant interference effect on the aerodynamic characteristics of aircraft, so it is necessary to do relevant research on the interference of tunnel wall. According to the characteristics of wind tunnel experiment, the flow field of a standard model in unsupported state in medium-sized wind tunnel open test section, medium-sized wind tunnel closed test section, large-sized wind tunnel closed test section and free flow field is numerically simulated By analyzing the variation of lift coefficient interference and drag coefficient interference with angle of attack, the following conclusions are obtained: the lift interference laws of the two sizes of closed wind tunnels are similar, and the interference increases with the increase of angle of attack, reaching the maximum when the angle of attack is 60 degrees, and then decreasing after the angle of attack is 60 degrees, and the lift interference of the open wind tunnel is the smallest, and the maximum appears at the angle of attack of 80 degrees. The law of resistance interference in closed wind tunnels with two sizes is similar, and the interference amount increases with the increase of angle of attack, while the resistance interference amount in open wind tunnels is the smallest. The outer boundary of the main flow area in the upstream of the aircraft is basically the same at different angles of attack, and in the downstream of the aircraft, the outer boundary of the main flow area gradually increases with the increase of angle of attack.

Although linear address space is very simple, it forces a two-dimensional matrix to be stored using either row-major mode or column-major mode. These two modes lead to non-uniform memory access latencies because of their different data localities. A common solution to improve the efficiency of column-major accesses is matrix transposition. However, transposition not only introduces extra data transfer operations but also increases the memory space consumption which is used to hold the transposed matrix or columns. If the column-major accesses can be performed as efficient as the row-major accesses, the transposition will be unnecessary. This paper proposes an interleaving-mapping data layout. With this method, both the row-major and the column-major accesses can be implemented with customized read/write instructions. The interleaving-mapping data layout have been implemented in scratchpad memory. The experimental results show that interleaving-mapping scratchpad memory can provide 1.4x speedup on average with only 1.73% extra area overhead.

Decision tree and naive bayes algorithms as classification algorithms are largely implemented in data mining. They are widely used in various data types which present accuracy differences among each type. In this essay, principles of two algorithms are introduced, and a specific data type, a large discrete dataset with incompletely independent attributes is used as a model to train those two algorithms. Then, two algorithms are tested, and confusion matrix method is used to calculate the accuracy of algorithms. Finally, two algorithms are compared, and it is concluded that accuracy of decision trees is better with large discrete datasets and attribute correlation.

To address the issue of imbalanced complexity and precision in series models of YOLO, this paper proposes a lightweight object detection algorithm based on YOLOv5s as the baseline model, namely YOLOv5-SFD. Firstly, the PANet structure of YOLOv5s is simplified to the basic FPN, and a Locally Dense Connected Convolution (LDCC) is proposed to improve the problem of feature information loss during downsampling by cheap feature extraction and feature reuse. Secondly, the C3 module of the original model is improved based on LDCC and combined with the C3Ghost module to reduce the weight of the model. Finally, a bottleneck residual block, BLS (Bottleneck LDCC-SE), based on LDCC and Squeeze-and-Excitation attention is introduced to enhance the feature extraction capability of the network. On the dataset NEU-DET, compared to the baseline model, the precision, recall, F1 score, mAP@.5 and mAP@.5:.95 is 73.4%, 84.7%, 77%, 91.5% and 52.5%, respectively, which represents an improvement of 3.2%, 2.2%, 2%, 11.8%, and 4.5%. In addition, the number of parameters is reduced by approximately 47%, and the model size is only 55% of the baseline model. The experimental results demonstrate that YOLOv5-SFD has lower model complexity and higher detection accuracy, which meets the requirements of lightweight networks.

Due to its simplicity and fast convergence speed, the particle swarm optimization (PSO) algorithm has been applied in many fields to solve various optimization problems. An improved PSO algorithm based on stream computing and multiswarm is proposed to solve the problem of low efficiency and easy to fall into local optimal solution for the traditional PSO algorithm, and it is abbreviated as SCMS PSO. In order to solve the problem of premature convergence caused by multimodal functions, this algorithm adopts multi-swarm. In the initial stage, the swarm is divided into several swarms based on the particle characteristics, and each swarm evolves separately, improving the global search ability of particles. Based on stream computing technology, this algorithm executes particle swarm evolving in distributed parallel. The experimental results show that the proposed algorithm not only has a faster convergence rate and more effective global search capability under the unimodal function, but also achieves better experimental results under the complex and deceptive multimodal function, which confirms that stream computing and multi- swarm can effectively improve the performance of the PSO algorithm.

The FASTQ format is a common genome file format generated by sequencing, mainly composed of identifiers, bases and quality scores. The quality score is an indicator representing the error rate of a base, which corresponds to the base. Although the compression of bases has achieved remarkable developments in recent years, the compression of quality scores remains challenging. This paper mainly improves ACO, a compression algorithm for quality scores. The main content of the algorithm includes the following parts. In terms of traversal quality scores, we comprehensively consider the sequencing process and the principle of FASTQ file generation. Instead of using the traditional raster scanning order, we choose a more reasonable serpentine traversal order. Base changes are introduced as contextual information. The weighted average of the quality score sequence is used as the index of the statistical characteristics of the sequence to perform clustering to improve the coding performance.

In order to solve the practical problems such as large number of information sources, information redundancy and difficulty in confirming target in Multi-UAV cooperative reconnaissance scenarios, this paper designs a multi-UAV target information fusion method combining target re-identification with online clustering technology. A real-time target feature extraction network is constructed to visually characterize reconnaissance target. The target information is encoded by fusing the target's longitude, latitude, category, and visual features. Hamming distance and cosine distance are used to measure the similarity of multi-source target features. Based on the online high-dimensional data clustering algorithm, we build a multi-source target information fusion architecture under Multi-UAV cooperative reconnaissance. Finally, the algorithm is verified based on the actual flight data of drone swarm composed of three UAVs. The experimental results show that the proposed algorithm can effectively fuse multiple reconnaissance target information, reduce the target search and decision time, and improve the efficiency of reconnaissance operations.

The effective use of high technology in all sectors of society can achieve the purpose of technological empowerment and facilitate the development of education, medical, aerospace, biomedical and other fields to overcome the difficulties. The rational use of advanced technology to achieve business development goals is one of the issues that all sectors are concerned about and must study. Artificial intelligence is a technology developed based on computer technology, through which computers can imitate human consciousness and improve their "cognitive" ability, enabling them to deal with various situations in a continuous cognitive manner. The article explores the application strategies of artificial intelligence in computer networks with a view to providing references for effective optimization of computer networks.

In this paper, an interval optimization method for uncertain structures is proposed. To obtain the lightest structure, a combination of interval mathematics and optimization algorithm are utilized. The uncertain parameters of the structure are described as interval parameters, and the upper and lower bounds of displacement constraints are derived by transforming the uncertain structural equations with the interval expansion. In order to obtain the optimal solution of the uncertain structure, the Particle Swarm Optimization (PSO) with dynamic time-varying inertia parameters is applied to optimize the structure, and effectively integrate it with the interval method. Based on the methodology, the optimal solution intervals of the structure can be obtained. This method provides a new approach for optimal design of uncertain structures.

The research on optimization algorithms has kept pace with the times, and one of the hot spots that has attracted much attention is the improvement of Particle Swarm Optimization algorithm (PSO) in the field of swarm intelligence. In this paper, a hybrid PSO algorithm using Adaptive Learning strategy (ALPSO) is introduced, and a hybrid PSO based on Arena Competitive Learning strategy (ACLPSO) is proposed from the perspective of search space and candidate particle generation, effectively enriches population diversity, expands the initial particle swarm search space, and improves convergence speed and accuracy.

Swarm intelligence has been the research focus of optimization problems, but there is no specific description and comparison of the most suitable application dimensions and scenarios for the recent swarm intelligence optimization algorithm. In this paper, the newly emerging swarm intelligence optimization algorithm in recent years is selected for numerical simulation experiments, and qualitative and statistical analysis is carried out, and the optimal optimization algorithm in different dimensions and test set is obtained, which provides faster convergence speed and better convergence algorithm for parameter optimization researchers under different situations.

With the development of technologies such as the Internet of IOT, data mining is becoming mature, and enterprise management will adopt these new technologies, and the safety management information platform can reach the bottom distributed hardware directly from the top decision-making layer, relying on the timely collection of data and real-time data analysis. Making use of intelligent analysis is of great value in monitoring the safety production of enterprises. In this thesis, the key points of the is the safety monitoring of Electric Power System. Using the characteristics of industrial Internet of Things (IoT), such as "connecting things", data collection and intelligent analysis, will improve the efficiency of Electric Power System. The enterprises will generate huge IoT monitoring data every day, which has high reference value for production process analysis and production decision making. storing and analyzing these IoT data in real time is a thorny problem. About the above problems, this paper proposes to use industrial IoT and big data mining technology to achieve monitoring and monitoring resource integration, real-time analysis and early warning.

Intersections are one of the significant factors influencing the overall traffic efficiency of road sections. This paper proposes an optimization model of signal light timing, with Travel Time Index (TTI), vehicle delay time, and queue length as optimization goals to alleviate the pressure on road traffic and enhance the traffic efficiency of intersections. Morever, a traffic signal timing control method for a single intersection is designed in this article. The proposed method realizes individual update and target optimization according to the mathematical characteristics and modifications of the sine function and cosine function, which considerably improves the timing optimization efficiency. VISSIM simulation proves that our suggested method is quite competitive. Experiments demonstrate that the average maximum queue length of the road is decrease by 23.01%, and the average waiting time is reduced by 41.86%.

Automatic grammatical correction (GEC) is one of the more difficult tasks in syntactic analysis for natural language processing. Non-native speakers often have difficulty understanding grammatical nuances, and grammatical correction in natural language contains grammatical errors and collocation errors. Compared with statistical classification-based error correction models, deep classification error correction models are a big improvement, but deep classification grammar error correction models can only correct for closed-form class errors. To address the above problems, this paper explores the automatic English grammar error correction algorithm from two ideas: using deep classification models and treating the grammar error correction problem as machine translation, and locating grammatical errors by building encoder-decoder sequence-to-sequence (seq2seq) models to understand the semantic and word expression differences between incorrect and correct sentences in order to improve the operational efficiency of model training and the basic ability to correct errors. This study uses the TensorFlow framework for the empirical validation of the used models to test the effectiveness of using the seq2seq error correction algorithm model with an attention mechanism.

To address the complex multi-tasking application scenarios and unstable network topology of micro Unmanned Aerial Vehicle (UAV) clusters, this paper proposes a clustering algorithm MT-WCA. Firstly, the UAV nodes calculate their weight based on average connectivity difference, transmission capability, relative mobility and energy consumption. Then, under the constraint of cluster membership, the most suitable cluster head is elected to prolong the network's integrity time as much as possible. MT-WCA partitions task clusters rationally based on task objective information in the scenario, enabling the UAV cluster to complete tasks more effectively. The paper simulates multi-tasking work scenarios by MATLAB and compares the integrity time of UAV networks using different algorithms.

To address the issue of current target detection models struggling to detect small vehicles and complex depth models being ill-suited for embedded platforms with limited computing power and memory, NWD and IoU are combined for tiny object detection, and the lightweight backbone network EfficientFormer is used to replace the YOLOv5 model backbone, which finally improves the tiny object detection effect while reducing the model parameters to achieve the lightweight depth model. The experimental results show that on the actual data set collected from the project, the mAP reaches 82.1%, and the number of parameters is reduced by 8.5MB.

To solve the problem of traditional A* in terms of high memory consumption, long calculation time, excessive turning angles and insufficient computing resources for mobile robots that require high-performance algorithms in large and complex scenes, this paper proposes an improved A* algorithm based on the jump point search algorithm. The algorithm selects a small number of key jump points as path nodes to generate the optimal path, reducing the operations of traditional A* algorithms on a large number of unnecessary nodes and thus reducing the demand for computing resources and improving algorithm performance. To verify the advantages of the improved A* algorithm, this paper conducted simulation experiments on 2D grid maps of different sizes. The simulation results show that compared with traditional A* algorithms, the improved A* algorithm requires fewer expanded nodes in the pathfinding process, has faster pathfinding speed, and has more significant performance improvement effects with large complex map.

In order to reduce the carbon emissions of thermal power in the combined dispatching system and ensure the safety and stability of power grid operation, a combined dispatching model of wind, photovoltaic, hydropower, thermal power and storage with the largest economic benefits and the smallest thermal power output fluctuations was constructed. At the same time, based on the standard NSGA-II algorithm, an improved NSGA-II algorithm based on adaptive arithmetic crossover operator and good point set initialization is proposed. Finally, the actual data of A and B in East China were selected for example analysis. The results show that compared with the standard NSGA-II algorithm, the improved NSGA-II algorithm can search for a more low-carbon economic dispatch strategy.

Space debris detection is of great significance for maintaining the safety of space activities and the sustainable use of space. Visible light detection has the advantages of long detection distance, high image resolution, simultaneous detection of multiple targets and small size, it is one of the important means of space target observation. In the case of long-exposure shooting of ground-based or space-based optical observation platforms, whether in target tracking mode or star gaze mode, there are large morphological differences between stars and space debris in the obtained star image. In order to detect dim small space debris more accurately, in this paper, we propose a new salient small target detection network which is named SD-Net in grayscale images (Space debris detection network). Specifically, we designed a new bi-directional attention aggregation module and used it to generate a weight map as auxiliary information to enhance the network's ability to learn object morphological information. And in order to further enhance the detection ability and robustness of the network, we add a multi-scale feature fusion module and use it for the final step of prediction at the end of network. Experiments result show that our proposed algorithm can detect a small amount of space debris well and is competitive in all evaluation indicators.

Given the large number of deployed air microstates but their low accuracy, it is not possible to accurately provide the true values of air pollutants. A data correction model (F-LSTM) combining Deep Neural Network (DNN) and Short-Term Memory neural network (LSTM) was proposed, and Particle Swarm Optimization (PSO) was added to obtain the true concentration values of PM2.5, PM10, nitrogen dioxide, sulfur dioxide, carbon monoxide and ozone. Use data from five air microstates in Zhengzhou High tech Zone in 2021 for validation. The results show that compared with LSTM and F-LSTM, the root mean square error and absolute mean error of the improved hybrid model (PSO-FLSTM) of particle swarm optimization are significantly reduced.

Most recommendation algorithms extract interest information from user behavior, but due to the rapid changes in user interest, freedom of user behavior, and poor rules of behavior data, the recommendation results are not ideal. Based on the above issues, combined with the strong regularity and high information accuracy of feature labels, a user behavior recommendation algorithm based on fused film feature labels is proposed, which introduces film feature labels and improves similarity measurement methods based on feature label weights. This algorithm establishes a user movie feature label matrix model based on movie labels in user behavior, constructs a new user preference feature model, and establishes a new similarity measurement method based on label weights to recommend matching movies based on user preference features. The experiment showed that on the Movielens dataset, the user coverage rate of the fusion film label algorithm reached 86.52%, and the accuracy and recall rate of recommendation results reached 58.89% and 35.59%, respectively, effectively improving the algorithm performance.

This paper proposes an improved genetic algorithm (IGA) to address the slow and unstable convergence speed of traditional genetic algorithms for solving traveling salesman problems. This algorithm optimizes the initial population through neighborhood search algorithms, designs an adaptive crossover and mutation probability, incorporates the Metropolis criterion to accept inferior solutions with a certain probability, improves the ability to jump out of local optima, and adds reversal operations to enhance local search ability and accelerate population convergence. Using MATLAB, IGA and five other algorithms were tested in the TSPLIB database. The simulation results showed that this algorithm has certain advantages in convergence speed and solution accuracy in small and medium-sized TSP problems.

Due to the low local optimization accuracy, long search time, and susceptibility to local optima in traditional genetic algorithms, their applications in path planning and other aspects are not very widespread. Therefore, this paper proposes a new improved genetic algorithm, which improves the crossover mutation operator on the basis of traditional genetic algorithms, making the algorithm effectively avoid falling into local optima problems and has stronger search ability, simultaneously incorporating the idea of simulated annealing to obtain a new population, it has stronger adaptability and global optimization ability. The experimental results of simulation research on the improved genetic algorithm show that the algorithm has relatively good superiority and feasibility.

Real time positioning and map construction are the basic technologies used by global navigation satellite systems when encountering environmental map information that cannot obtain individual positions. They are a fundamental capability required for many mobile robot navigation tasks. This article utilizes the constant speed model to apply it to the adjacent relationship logic of keyframes and decouples and optimizes visual inertia to reduce the computational power of the SLAM framework and achieve lightweight algorithm.

The immutability of blockchain can ensure the security and veracity of fresh food information, but in the process of data collection and processing by edge nodes and servers, a large quantity of data that must be uploaded to the blockchain network for storage will be generated quickly. This paper designs and proposes the edge cloud data collaboration technology of message queue, which uses RabbitMQ message middleware to provide cache space for the data waiting to be uploaded to the storage and achieves batch uploading by matching the throughput of the data accepted by the blockchain network, in order to address the problem that the data uploading speed does not match the data generation speed and thus causes data backlog. On the basis of an information tracing architecture that combines peripheral cloud collaboration and message queue technology, smart contracts were developed to submit authenticated data to the blockchain and IPFS. The prototype verifies the viability of the system and ensures the safety and dependability of the storage of fresh food data traceability information.

With the increasing variety of position sensors involved in navigation and positioning technology, the advantages and disadvantages of each sensor collecting data coexist, and relying solely on a single sensor to provide information cannot meet the requirements of actual navigation. This paper uses Kalman filtering technology to preprocess the position information collected by GNSS sensors, visual sensors, and inertial sensors, and calculates relative error and root mean square error to judge the confidence and accuracy of the pre-processing. Then, the filtered data is fused by introducing an optimal weighted fusion estimation algorithm, and the simulation results of the algorithm are analyzed and discussed.

The mainstream deep learning spoken language evaluation algorithms are based on speech recognition to perform mispronunciation determination. In the process of speech recognition, Deep Neural Network-Hidden Markov Model (DNN-HMM) has better performance than Gaussian Mixture Model-Hidden Markov Model (GMM-HMM). In this paper, the structure of speech recognition system is analyzed and the current mainstream English speaking evaluation algorithms are analyzed, and it is concluded that the current mainstream evaluation algorithms are basically not end-to-end and have low evaluation metrics. Therefore, this paper designs an end-to-end speech evaluation model based on the theory of deep learning, and conducts experiments for this model under the TIMIT speech dataset to verify and analyze the advantages of this paper's model in temporal modeling, which shows that the speech recognition model designed in this paper has better performance in spoken language evaluation.

Reducing logistics costs in the supply chain has been a key concern of major enterprises. Meanwhile, reducing the delivery cost of the supply chain is a key component of supply chain cost management. Therefore, this article mainly focuses on the cost management of agricultural enterprise supply chain delivery and achieves the goal of reducing logistics costs by optimizing the specific process of supply chain delivery. Firstly, a multi-objective mathematical model was established by comprehensively considering multiple objectives such as delivery cost targets and carbon emission targets in the delivery process of agricultural products. Furthermore, the delivery quality of agricultural products during the delivery process was ensured through time window constraints, and corresponding constraint conditions were established based on the actual scenarios during the agricultural product delivery process. On this basis, an Improved Artificial Bee Colony algorithm (IABC), a heuristic algorithm, was designed. And simulation experiments were conducted. The simulation results showed that the algorithm designed in the article can be effectively applied to the cost management problem of agricultural enterprises.

The ability of the human autonomic nervous system to regulate can be assessed and judged by HRV analysis. The purpose of this paper is to examine the HRV of college students in the quiet condition and the heart rate variation of the recovery process following long-distance running, and heart rate change from rock music to a light music environment were compared and analyzed. By studying the control group consisting of collegiate and non-varsity sports teams as objects of investigation, and the effects of exercise and noise on HRV were assessed. For the purposes of this paper, the ECGs of experimenters in different states were collected, and the R-R interval was extracted as the primary research data. First, the HRV parameters MHP, HR, SDNN, RMSSD, pNN50 in the time domain were calculated using the R-R interval and then the HRV in the frequency domain was analyzed by scatterplot analysis. Based on these results, the average specificity of the five indicators was 88.74%, and the sensitivity was 76.54%. In conclusion, this study demonstrates that HRV of collegiate sports teams in the process of sport recovery and music slowing is stronger than that of non-varsity sports teams. And exercise may improve the balance of sympathetic and parasympathetic nerves in the human body.

Parking space detection is an important part of an automatic parking system, and its detection performance determines the final parking effect. In this paper, a yolov8-based parking space detection algorithm is used, and the speed of detection is greatly improved because we propose an end-to-end network. The main research work is as follows: a yolov8 based instance segmentation model is used to process the image, the result of the instance segmentation is converted into a gray scale map and then feature contour points are extracted, and then a polygon fitting algorithm is used to correctly detect parking spaces in the map. The detection using this method is better, but the speed is reduced. The network structure was also modified and a new loss function was invoked to further improve the accuracy of the model.

In order to realize the rapid dispersion of a large number of agents into a specified space, a multi-agent formation identical particle diffusion algorithm is proposed. By simulating the cell division process, an identical particle diffusion algorithm is designed, so that the number of particle swarm becomes three times that before the transformation after each change, and can cover a large area. The simulation results show that the algorithm can be spread over multiple regions in a short time. The algorithm can be widely used in the fields of computer technology, science, physics and transportation. It can also be used in the cluster control of unmanned aerial vehicles. In particular, this algorithm can be applied to quickly disperse a large number of agents into a large area.

In the development of modern industry and the harmonious coexistence of human society and nature, fires often occur due to various accidents and negligence. Such accidents are extremely detrimental to social development. Compared to traditional smoke detectors, video-based smoke detection can achieve non-contact detection without the need for sensor installation, reducing equipment costs and installation difficulties, making it an important development direction. Video smoke detection based on SSIM (Structure Similarity Index Measure) has the advantage of being simple and easy to implement and can perform well in some simple scenarios. However, it may have a high rate of false positives in complex scenes. Therefore, in this paper, the algorithm is improved by using texture features for further detection, improving the accuracy of detection and increasing the innovation of the algorithm. The experiments show that the improved algorithm achieves good results.

A resource allocation strategy based on the improved Marine Predators Algorithm and KM algorithm is proposed to address the interference problem generated by spectrum multiplex in the coexistence scenario of Train to Train (T2T) communication and Train to Ground (T2G) communication in urban rail transit system while ensuring the user communication quality. First, the throughput optimization model under multiple constraints is developed and decomposed into two sub-problems of optimal power allocation and Channel Multiplex matching. Second, considering the problems of premature convergence and easy falling into the local optimum of the original MPA algorithm, a reverse learning strategy is used to optimize the initial population, improve the step size parameter and introduce the search balance parameter to improve the performance of the algorithm, and allocate optimize transmission power for T2T users and T2G users. Finally, the KM algorithm is used to find the best Channel Multiplex combination. Simulation results show that this algorithm improves the overall system throughput by 5.9% and 7.1% compared with Marine Predators Algorithm and Gray wolf optimization algorithm, significantly improving overall system communication performance.

The point cloud data obtained by a 3D laser scanner is known for its high density, redundancy, and significant noise, which can result in a slow 3D reconstruction process and an imprecise reconstruction surface. To address these issues, a method for rapidly creating a 3D point cloud model based on 3D laser scanning was proposed. This approach involves removing outlier point clouds using a statistical filtering algorithm, downsampling the data using an improved octree voxel filtering algorithm, smoothing the point cloud using the moving least squares method and resampling it. Finally, the improved greedy projection triangulation algorithm is used for surface reconstruction. The experimental results show that the method proposed in this paper can improve the efficiency of model reconstruction, and the reconstructed model has smooth surface, accurate structure and improved precision.

For the real road environment, the proportion of traffic sign targets is small and not easy to be detected, and the recognition background is complex and changeable, resulting in the problems of low detection precision and poor robustness. In this paper, a traffic sign detection algorithm based on improved Yolov5-PB is proposed. The algorithm uses the ParC-Net as the backbone feature extraction network to fully extract the target feature information. The BiFPN structure is adopted to enhance the feature fusion ability of the network for multi-scale traffic signs. Simulation is carried out on TT100K and GTSDB traffic sign datasets, and the results show that the mAP@0.5 of Yolov5-PB algorithm reaches 85.6% and 72.2%, respectively, which are 3.1% and 19.2% higher than Yolov5. The precision reaches 85.3% and 71.6%, respectively, which are 0.3% and 21.8% higher than Yolov5. Furthermore, compared with the current mainstream target detection algorithms, the proposed algorithm can achieve better detection precision and meet the requirements of real-time detection, and has better robustness.

This paper presents a robust point cloud optimization algorithm based on nonadjacent low-altitude remote sensing images. The proposed algorithm is designed to optimize the performance of point cloud generation in accuracy and efficiency. In order to accelerate the process of stereo matching and compute the coordinate of object point accurately, a pair of nonadjacent images is employed to lengthen the epipolar line of the image pair. Then a patch based Least Square Matching (LSM) method is utilized to search the optimal matching pixels and compute the coordinates of corresponding object points in 3D space. Comparison studies and experimental results in point cloud generation about low altitude remote sensing images have proved the effectiveness of the proposed algorithm.

UAV swarms have crucial applications in modern military, geological exploration, and 5G/6G communication fields. As communication nodes, drones frequently exchange wireless data with other drones, and data privacy protection is currently one of the most urgent research topics. Based on this, this article proposes an efficient isomorphic federated learning algorithm for unmanned aerial vehicle clusters. First, set the adaptive loss differential adaptive parameter, with the initial value set to a floating point number not greater than 0.01, then activate it with the Tanh function, participate in the training as part of the loss function during the training process, and optimize it by gradient descent. When the UAV node uploads the gradient to the wingman, sum the adaptive loss differential parameter with the gradient matrix as a disturbance term. Then, based on the dynamic confidence matrix, high-quality drones are selected to participate in the gradient security aggregation of drones, achieving the selection of high-quality drones. The built-in global neural network of the drone transmits shared parameters indiscriminately to each drone through broadcast to achieve updates to the shared parameters. The comparative experiments of our algorithm on the Fashion and Cifar10 datasets show that our algorithm has higher accuracy, with the highest accuracy improvement of 4.42% on the Mnist dataset and 8.22% on the Cifar10 dataset.

Text semantic similarity computation is a fundamental problem in the field of natural language processing. In recent years, text semantic similarity algorithms based on deep learning have become the mainstream research paradigm, but they suffer from the problem of insufficient understanding of text semantics and thus unclear interpretation of computational results. In this paper, we propose a text semantic encoding model combining frame semantic theory and deep learning, which enriches the semantic representation within sentences by making full use of the frame, frame elements, lexical units, and inter-frame relations in the frame semantic knowledge base and combining them with the Bert model, and interacts with the semantics between sentences through Siamese transformer encoders. The experiments are validated on MSRParaphraseCorpus and quora-question-pairs, and the results show that the model proposed in this paper outperforms similar models in terms of F1 values.

With the explosive growth of internet information and the increasing demand for service quality from users, personalized recommendations have become increasingly important. This article proposes a recommendation algorithm based on deep learning models. Utilize the powerful learning ability of deep learning models to mine more hidden information in data and fully extract features of users and projects. At the same time, in order to solve the problem of data sparsity and the dynamic change of user preferences over time, in the prediction generation stage, user preferences are integrated and updated in a timely manner. The experimental results on the dataset show that the algorithm has achieved good results in prediction accuracy and recommendation quality.

Many industry fields like aerospace and automotive require deterministic network transmissions. TSN (Time-Sensitive Networking), a data link layer technology that can provide low latency and high bandwidth, is gradually gaining attention. IEEE 802.1Qbv, as one of the most important standards in TSN, provides deterministic time slots for the transmission of different flows through GCL (Gate Control List). In this paper, a heuristic algorithm is proposed to address the problems of GCL solving complexity and large GCL cycle for large-scale flows. All flows are divided into a new period set by pre-processing and then scheduled according to the new period set. The engineering application conditions of this algorithm are also considered. The experimental results show that it is feasible to quickly compute GCL for 100000 flows using the proposed algorithm, and the GCL cycle is smaller than computed using the original way.

With the improvement of medical information security and sharing requirements, the blockchain Electronic Medical Record (EMR) platform has become an effective choice to solve the problem of securely storing and sharing medical information. The consensus mechanism is the key to EMR blockchain technology. Raft consensus is the most widely used non-Byzantine fault-tolerant consensus algorithm, which has the disadvantage that election disputes occur easily, and it cannot adapt to malicious nodes. Therefore, this paper proposes a raft consensus algorithm based on a credit evaluation model (CO-Raft) with the OC-SVM algorithm to filter out Byzantine nodes in the chain and build a dynamic node evaluation model based on the credit value to improve the speed of leader selection. Experiments show that the CO-Raft consensus mechanism in this paper provides better security and higher throughput in the EMR blockchain.

In order to ensure the smooth entry of large stowage units, reduce manual assistance, improve loading efficiency and realize automatic loading, this paper takes the large unit devices as the research object, and puts forward the problem of realizing the optimal trajectory of the large unit loading devices of aircraft cargo compartment based on the improved adaptive genetic algorithm. A trajectory planning model is established with the constraint conditions of the left side of the hatch door, the right side of the hatch door and no collision of the hatch body, and the goal of the shortest loading trajectory. An adaptive genetic algorithm was designed to solve the model. The crossover probability and mutation probability were automatically adjusted with individual fitness through genetic control to improve the convergence speed and get the optimal solution faster. The simulation verifies the feasibility of the proposed model and algorithm, which has reference significance for the loading of large ULDs.

Recently, with the development of new energy vehicles, there has been an increasing demand for new energy batteries. During the production process of batteries, laser welding is an important technology, which can produce defects inevitably. So, the defects detection of laser welding is absolutely necessary to ensure the quality of the batteries. In this paper, we proposed an improved YOLOv7 algorithm for laser welding defect detection of battery pole. The algorithm adopts Conv2former to improve the model's capability to extract contextual information and introduces the CBAM attention mechanism to extract key information. The loss function is also modified to WIoU to further improve the accuracy of the model. The experiment results on the battery pole dataset show that the improved YOLOv7 achieved a 3.1% increase in map@0.5, reaching 0.921, and a 2.1% increase in map@0.5:0.95, reaching 0.682 compared to the original YOLOv7.

This paper studies a Permutation Flow Shop Scheduling Problem (PFSSP) considering maintenance activities and energy consumption, in which machine failures obey Weibull distribution. In case of an unexpected failure, a Corrective Maintenance (CM) policy is triggered automatically. Moreover, Preventive Maintenance (PM) as a proactive maintenance strategy is introduced to improve deteriorating effects and reduce the probability of failures. Then, a multi-objective production-maintenance joint optimization model is proposed. Compared to the traditional PFSSP, this joint model is more complex. To solve it, a Reinforcement Learning (RL)-based solution approach is designed with specific state sets, action sets and reward functions. Numerical studies show that the application of RL in integrated scheduling has high solution performance by comparing with seven heuristic rules.

As an important part of power grid emergency management, power communication fault emergency repair scheduling can provide decision support for emergency management. Therefore, based on the current situation of emergency repair resource allocation of power communication, a multi-point fault emergency repair scheduling algorithm considering multi group cooperation is proposed, and a joint optimization model of power communication fault emergency repair considering resource allocation and multi team cooperation mechanism is established. The blackboard information processing mode is designed, which is composed of initial layer, adaptive measure optimization layer and emergency repair plan optimization layer. It can quickly give the emergency repair resource grouping scheme and emergency repair plan after multiple failures of power communication, so as to reduce the economic loss of power failure. Simulation results show the effectiveness of the proposed method.

Problems such as low trust, opaque data and data leakage are easily found in data sharing of evidential testing among different power engineering departments responsible for quality management. In this regard, this paper proposed several schemes for a secure data sharing in the power engineering. In the schemes, each department forms an allied blockchain, and the generated data for testing are stored on-chain and off-chain. To ensure data security and sharing, the testing data are encrypted into ciphertext, and Ciphertext-Policy Attribute-Based Encryption (CP-ABE) is used to realize cross-unit sharing of ciphertext data. The proposed schemes can effectively solve the problems existing in data provenance, data sharing and privacy protection in an engineering data testing and can promote the application of blockchain in this field.

With the rapid development of quantum computing technology, data sharing systems in cloud storage may suffer from quantum computing attacks. Recently, many researchers have proposed encryption schemes to achieve quantum attack resistance and fine-grained data access control. However, single point of failure and secure attribute revocation are two critical problems for quantum-resistant encryption scheme in cloud storage environments. In order to deal with these problems, in this paper, we propose a revocable multi-authority attribute-based encryption from lattice (RMA-ABE-L) in cloud storage. Firstly, RMA-ABE-L scheme can achieve fine-grained data access control with quantum computing attack resistance. We design efficient anti-quantum algorithms based on R-LWE problem and merge them with ciphertext-policy attribute-based encryption (CP-ABE) structure. Secondly, RMA-ABE-L scheme can solve the single point of failure problem. We design the multi-authority structure to so that the compromise of some authorities does not affect the functioning of the whole data sharing system. Moreover, RMA-ABE-L scheme can realize secure attribute revocation. We design a secure attribute revocation mechanism and set special identity construction. The performance analysis shows that RMA-ABE-L scheme is more efficient than other related schemes.

Ultra-wide band (UWB) technology has become a mainstream technology of high-precision IoT positioning. Due to its large frequency band and narrow pulse characteristics, UWB signals can well cope with multipath effects, and are currently widely used in personnel positioning in factories, goods positioning in warehouses and other scenarios. However, due to the non-line of sight (NLOS) problem, its positioning performance will be greatly reduced in the actual scene. The existing schemes are mainly solved by threshold method and feature method, but the effect is not ideal. This paper proposes a unified UWB indoor positioning framework based on deep aggregation network, called UANet. In UANet, phase and amplitude information are obtained by processing channel impulse response (CIR), and then a deep aggregation network integrates CNN and Transformer to identify NLOS scenes. Finally, weighted least squares method (WLS) is applied to complete high-precision positioning of indoor scenes. We conduct extensive experiments in real scenes, and the average positioning accuracy of different depth learning models and traditional UWB positioning methods is compared. The experimental results show that compared with traditional methods, the proposed UANet achieves an average positioning error improvement of 6.4 cm

Proxy re-encryption plays an important role for data security in cloud computing and big data, but the traditional re-encryption scheme based on the classical number theory problem cannot resist quantum attack, and the lattice-based cryptosystem has been proved to have the ability to resist quantum attack, so the lattice-based re-encryption scheme has always been a research hotspot. In view of the problem that collusion-resistance and multi-hop cannot be satisfied at the same time in existing schemes, this paper proposes a new identity-based proxy re-encryption scheme on lattice. Firstly, the scheme uses trapdoors as private keys to achieve one-way anti-collusion. In addition, we construct a variant of Learning with Errors (LWE), which uses a convert algorithm to transform the user matrix and replace the vector with a uniform matrix, so that the scheme can meet the requirements of multi-hop and encrypt multiple bits of messages at one time. Theoretical analysis and experimental simulation prove that the security of the new scheme can be reduced to a decision LWE problem, which can reach the IND-sID-CPA security in the standard model and can improve the encryption efficiency while ensuring the security.

Unlike long texts, short texts can cause the problem of feature sparsity due to their short length. Although the existing deep learning-based methods alleviate this problem, the extracted semantic features are inevitably redundant, and the relationships between features are not fully considered when performing feature fusion, which in turn makes the semantics ambiguous and classification very difficult. To address the multi-feature fusion problem, we propose a Short Text Classification Model Based on Dynamic Routing and CNN with Attention Mechanism (DCAN). Firstly, we use Convolutional Neural Network (CNN) to extract text features with different granularity to enrich the semantic representation; secondly, we use attention mechanism and residual structure to fuse features with different granularity to obtain features with contextual semantic relationships, and then them into dynamic routing to obtain the probability distribution of short texts. We conducted experiments on four datasets, including AG News, MR, TREC, and SST-2. The experimental results show that DCAN has higher classification accuracy compared with most of the currently popular models.

Aiming at the specific requirements of the comprehensive risk assessment system for the whole mine, such as strict quantitative basis, evaluation result percentage system, automatic and real-time evaluation process, the analytic hierarchy process based on neutral reference objects is adopted to conduct comprehensive risk assessment for the whole mine. The evaluation system constructed by this method does not require manual intervention during operation and is easy to integrate various existing evaluation methods for comprehensive evaluation. On this basis, a comprehensive risk evaluation index system and evaluation model for the while mine were established, including three primary indicators of disaster risk, safety compliance, and other safety factors. Two methods for handling missing data during safety evaluation were proposed, and a brief explanation was given of the major hidden danger veto system during the evaluation process. Finally, the logical structure, business processes, risk visualization methods, and specific implementation technical routes of the comprehensive risk assessment system for the whole mine are introduced, and application demonstrations have been conducted in multiple practical projects.

As e-commerce continues to grow in popularity, the study of abnormal data in e-commerce orders is increasingly important. By analyzing abnormal data in e-commerce orders, one can uncover problems and trends within the order data, ultimately leading to improved operational efficiency. However, many abnormal detection models face issues such as overfitting, excessive dependence on sample size, and the neglect of other important features. To address these issues, this study proposes a novel approach that combines the ensemble model with MLP. Specifically, this approach leverages the strengths of Random Forest, GBDT, and XGBoost, which serve as the first-layer feature extractors. Samples that are misclassified in the first layer are discarded, and the remaining training samples are input into the second-layer MLP. This approach enhances the model's generalization ability, strengthens its global modeling capability to a certain extent, and mitigates the problem of excessive dependence on samples. The dataset from Gitee validates that this two-layer model can effectively extract useful information contained in the data, and its accuracy surpasses that of classical ensemble models.

Due to the problems of feature redundancy and high dimensionality in Software-Defined Networking (SDN) intrusion detection, the original Seagull Optimization Algorithm (SOA) appears to have lack of exploration capability and tends to fall into local optimum problems. A hybrid strategy Improved Seagull Optimization Algorithm (ISOA) is proposed to solve these problems. Firstly, a cubic chaotic mapping strategy is used to initialize the population, laying the foundation for the global search for an optimum. Secondly, a Cauchy-Gauss variation strategy is introduced to improve the search efficiency of the algorithm. Finally, a simulated annealing algorithm is combined with a certain probability to accept a solution that is worse than the current one, jumping out of the local optimum. The ISOA was applied to the InSDN dataset to verify the intrusion detection performance of the model. Comparing with the original SOA and six other classical algorithm models, the data dimensionality was significantly reduced and effectively improving the performance of SDN intrusion detection.

With the development of Live-Virtual-Constructive (LVC) simulation technology, numerous LVC simulation resources have been developed. To construct more large-scale LVC simulation, it is necessary to integrate existing simulation resources. Due to the heterogeneity of LVC resources, the interoperability between heterogeneous simulation resources becomes a major concern. This paper proposes a common gateway in multi-architecture LVC simulation environment that converts messages between different simulation architectures and achieves a level of semantic interoperability. The gateway provides a collection of protocol-independent interfaces for simulation systems to access co-simulation conveniently. To maintain timing consistency in heterogeneous simulation environments, a time synchronization method is employed to coordinate logical time and real time. Experimental results show that the scheme is flexible and can maintain data transmission latency at a millisecond level in LVC simulation environment, which includes High Level Architecture (HLA), Data Distribution Service (DDS), and Distributed Interactive Simulation (DIS).

In the last few years, the Internet of Things (IoT) has become important part of our lives and its influence is expected to exponentially increase in the next years. Computer systems and applications are improved day by day and with the advancement in such area it gives birth to new cyber-attacks. Man in the Middle attacks (MITM) are one of those attacks. An attack where an outsider or third party enters in between two online users, where both of the users are unaware of it. Mainly it is knowing as a protocol to an unauthorized user within the system who can access as well as change the information of the system without leaving any trace to the existing users. Cloud, edge and fog computing added more flavors in terms of better throughput and low latency in ease of accessing data. Edge computing refers to a distributed open platform which helps off-loading tasks and processing of cloud at edge of the network. The edge serves as a bridge between the digital and physical worlds making gateways, systems and services smart and efficient. Due to the exponential growth of IoT produced data, its complexity is also expected to increase and raises need for innovation for improved security in Mobile Edge Computing (MEC) environment. This research is conducted to propose a lightweight Blockchain solution using private/ public keys and employing shading for off-loading computation intensive tasks at MEC. Inter-Planetary File System (IPFS) database has been used to offload computational task from edge servers. Proposed technique is implemented using Java Script and Node.JS and simulated through Visual Studio Code and results are compared with other lightweight encryption algorithms proposed for IoT.

To address the problems of network intrusion data imbalance and prediction accuracy, a network intrusion detection model based on SMOTEENN (Synthetic Minority Oversampling Technique Edited Nearest Neighbor) and improved CatBoost algorithm is proposed, and SMOTEENN sampling is used to solve the data imbalance problem from the data level. The original CatBoost algorithm loss function is replaced with focal loss function to enhance the classification performance of CatBoost algorithm from the level of sample classification difficulty. And the Bayesian optimization algorithm is used to realize the CatBoost hyperparameter combination seeking. The network intrusion dataset NSL-KDD (National Security Letter Knowledge Discovery in Database) is input to the improved model and compared with other similar algorithms. The results show that the proposed method in this paper is optimal in terms of accuracy, F1-sacore, and AP value, To provide new ideas for network intrusion algorithm research.

The implementation of Software-Defined Networking (SDN) as a new network architecture brings about a variety of novel characteristics and application scenarios, decreasing network development redundancy, but it also raises serious security concerns. This study describes and develops the recurrent neural network-based LSTM-GRU detection algorithm for Distributed Denial-of-Service attacks (DDoS) assaults in SDN, which uses LSTM and GRU as the base model and stacks them to achieve high accuracy classification of traffic. According to tests, the detection algorithm outperforms conventional detection techniques, achieving accuracy levels of 99.53% and 98.78% on the CIC-IDS 2017 and CIC-DDoS 2019 datasets, respectively.

In recent years, more and more computer technology has been applied in industrial Internet, the vulnerability and security risk of industrial control systems are gradually increasing. network security issues have received much attention. To address the complex attacks and relatively poor deployable defense resources in the Industrial Internet, this paper combines industrial Internet attacks with signal game theory, proposes a moving target defense strategy selection method. The SDN-based industrial Internet attack and defense model is established, the core decision module and the moving target defense module are introduced to select and implement the best defense strategy respectively, so that the industrial Internet can achieve the best security defense strategy. The simulation results show that the moving target defense strategy selection method based on the signal game can effectively improve the defender's gain and achieve the best defense against network attacks under the industrial Internet.

This study ensures the safe operation of bridges by real-time recognition of concrete surface cracks on bridges. Traditional crack recognition methods are slow and have low accuracy. To address this issue, we propose an improved YOLOv5 (You Only Look Once version five) detection model based on Deformable Convolution (DCN), called YOLOv5-D. The baseline YOLOv5 model has been optimized in three aspects: First, deformable convolution replaces part of the standard convolution in the backbone network, improving the model's recognition accuracy. Second, the SE attention mechanism is added to adaptively recalibrate channel-wise feature responses, stimulating information features and thereby enhancing the shared underlying feature map representations. Third, the CIOU_Loss is proposed as the loss function, achieving a collaborative optimization of crack detection speed and accuracy, and meeting the deployment requirements of embedded terminals. The improved YOLOv5-D model achieves an average recognition rate (mAP) of 76.1% and a recognition speed of 54 FPS with a model size of 14.5 MB. Compared to the original YOLOv5 model, the average recognition rate is increased by approximately 5.9%, the parameter quantity is reduced by about 10%, and the detection speed remains essentially unchanged. In summary, the YOLOv5-D model achieves a better balance in terms of accuracy and speed, effectively locates cracks, and provides a new technology for early warning of bridge collapse disasters.

A chronic neurological condition called epilepsy affects about 50 million people globally. Although there has been a lot of study on deep neural networks to identify seizures, it is still difficult to precisely define the structural relations between electrodes in EEG signals. The difficulty involved may lead to less accurate prediction. We provide a novel method that combines structured learning methods with graph neural networks and incorporates graph convolution to provide explanations for detected anomalies in EEG signals in order to overcome this constraint. Our investigations on the CHB-MIT dataset show that our approach is more effective and efficient than baseline methods and conventional neural network models at identifying the onset of seizures.

With the improvement of renewable energy utilization rate, its proportion in power generation continues to increase. However, the uncertainty of renewable energy power generation has always been an urgent problem. Accurate forecast of short-term power load can assist modern power systems in accurately allocating the proportion of traditional and renewable energy generation. This is crucial for the safety and economic operation of the power grid. Power load forecasting can be seen as a multivariate time series forecasting problem. Most of existing studies have focused on predicting the next moment power load. However, a great number of decisions in power planning scenarios require predictive models that can provide a complete conditional distribution with richer information, namely probability distribution prediction. To address this limitation, we applied a multi-step probability prediction model based on recurrent neural networks on short-term power load forecasting. Experiments conducted on the annual power load of multiple regions in the United States and comparison with some classic models verify the superiority of the applied model. The results indicate that the model applied in the article has a promising predictive performance.

With the widespread application of the Internet of Things, millions of sensors, actuators, or embedded electronic devices are interconnected through the Internet. Monitoring a large number of IoT devices is a huge challenge, especially when sensor nodes are limited (such as limited CPU, memory, and batteries). The IoT device management protocol has become an important research field. However, currently, management protocols designed for the Internet of Things, such as CoAP and MQTT, do not provide security support. SNMP is a widely used protocol for monitoring systems, and its v3 version provides comprehensive security support. However, SNMPv3 uses complex encryption algorithms and cannot be applied to resource limited IoT devices. In order to address this issue, we propose a secure and lightweight implementation scheme for Internet of Things (IoT) device management based on the ZUC algorithm. This solution utilized the ZUC algorithm as its core and created authentication and encryption protocols in order to provide safe connection and control ofloT devices. We conducted simulation experiments and compared our scheme with the simple network management protocol to confirm its viability. The results show that, while maintaining security, the proposed solution greatly lowers computing and communication overhead, making it suitable for Internet of Things (IoT) scenarios.

The operation of distributed systems highly depends on reliable and continuous communication between nodes. The failure detection phase thus plays an important role in system unavailability. Previous studies have shown potentials for LSTM based failure detectors but the heavy computation cost remains a problem. We propose a technique to reduce the computation time of LSTM based FD by performing forecasting instead of making one prediction at a time (i.e. predicting multiple timestamps at once). Our method achieves better computation time compared to previous LSTM FDs but falls short of BFD. We discuss the heavy computation cost of back-propagation of LSTM and the high frequency nature of FD tasks and conclude that LSTM isn’t likely to be the best fit for Machine Learning based FD.

Deep learning based Single image super-resolution has achieved great success, but the computational cost is too high, making it unsuitable for real-time tasks on resource-constrained devices. In this paper, we propose Residual Local Blueprint Separable Network (RLBSN), which based on RLFN lightweight backbone network. The blueprint separable convolution was used to replace the traditional convolution, which can enhance the extraction of edge information, reduce the number of model parameters, and improve the running speed. The knowledge distillation concept is introduced to calculate loss function and the output of high-performance model is taken as ground-truth to improve the performance of the model. We design the joint loss function that combines the characteristics of various loss functions to generate high-quality super-resolution images. Inference time can be significantly accelerated by transferring post processing to the GPU.

With the continuous development of the new power system, the broadband and isolation requirements of the power communication network are further increased. It is urgent to evolve the traditional PTN packet transmission network into SPN sliced transmission network with slicing capability. This paper proposes two networking schemes of service segment deployment and PTN OverLay SPN. Comparing the two schemes from the aspects of security, reliability, implementation complexity, operation and maintenance complexity, the paper analyzes the application scenarios of the two schemes and verifies the feasibility of the schemes through the actual scene test.

Based on the data and requirements provided in question D of the 2022 Chinese College Student Mathematical Modeling Competition, this article conducted research on the transmission process plan of meteorological message information satellite communication, comprehensively analyzed the impact of various factors on message information satellite transmission, carefully sorted out the relationships between various observation stations, and established a cyclic model, probability model, and linear equation model, The process problems during transmission were solved using methods such as mean and MATLAB software. The mathematical model in this article can be promoted and used in the transmission of meteorological information in surveying.

The route planning of mapping Unmanned Aerial Vehicles (UAVs) such as image-oriented and tilt photography is not adjusted and optimized in combination with the current situation of acquisition objects, which may cause the risk of collision between UAVs and higher objects. At present, the general design is based on cesium tilt photography 3D model monomer, which does not solve the automatic regular 3D model contour. For the spatial information matching of distribution network equipment, due to the nonlinear coordinate offset adopted by the power grid GIS platform, there is an offset between the collected coordinates and the GIS storage coordinates, which cannot be matched. Therefore, it is very urgent to realize the automatic association between the collected information and the PMS account information. This paper analyzes, designs, and implements the route planning method and the three-dimensional model monomer system. In order to solve the problems of unreasonable route planning of UAV in distribution network grid, collision between UAV and higher objects, incomplete collected data and so on, the route planning method is designed. This method optimizes the route by using the existing three-dimensional data, spatial analysis, and collision detection on the GIS platform. The system is analyzed and verified in a real city.

With the development of the Internet and the advent of the digital era, the Internet has become an indispensable part of people's daily lives. However, along with the development of the network, there are more and more cyber-attacks, which not only pose a threat to personal privacy, but also have a huge impact on national security and social stability. Therefore, how to effectively protect our network security has become one of the urgent issues to be solved. In the traditional network security defense, a rule-based approach is usually used for monitoring and analysis, but this approach has certain limitations. Because the network environment is complex and ever-changing, and various types of attacks are emerging, it is difficult to detect and respond to new attacks in a timely manner in the traditional way. More advanced technical means need to be introduced to improve the level of network security assurance. Therefore, this paper proposes a computer vision-based network security detection system based on the summary of existing network security detection methods and the development status of the computer vision field. The main idea of this system is to use computer vision technology to realize the functions of automatic extraction, characterization and classification of network data, so as to improve the efficiency and accuracy of network security detection.

The challenge of detecting Human-Object Interaction (HOI) in photos is addressed in this study through the introduction of an effective Transformer approach. The Transformer decoder incorporates Spatially Modulated Co-Attention (SMCA) to preset the position of the target (human or object) in the image, narrow the search range of the query vector and accelerate the convergence of the model. In order to fuse multi-scale features and increase model recognition accuracy, at the same time, the intra-scale self-attention mechanism is introduced into the encoder as an auxiliary operator to fuse multi-scale features. The model adopts multiple Multilayer Perceptron (MLP) structures to locate and identify HOI instances. The loss function of this model is the combination of classification loss and bounding-box regression loss. The HICO-DET dataset had a detection accuracy of 27.19%mAP on average, while the V-COCO dataset had a detection accuracy of 53.2%mAP.

Intrusion Detection (ID), as an effective network security protection technology, is an important means to ensure network security. In order to solve the problem that traditional intrusion detection technology is difficult to adapt to the current complex and changeable new network attacks, this paper uses the Random Forest (RF) algorithm of machine learning algorithm to solve this problem. Random forest algorithm has good feature learning and classification ability, which can effectively improve the accuracy of intrusion detection system. However, when the characteristic values of the sample data are high, its time complexity will also be high. In this paper, K-means clustering algorithm and One-Rule algorithm are introduced to improve the random forest algorithm, improve its modeling and classification speed, and further improve the detection efficiency by means of hierarchical division. In this paper, the proposed KOne-RF algorithm is verified by using NSL-KDD data set. Experimental results show that compared with the traditional RF algorithm, KOne-RF algorithm has improved modeling time, detection classification and detection accuracy, and detection ability for different types of attack behavior, which verifies the reliability of KOne-RF algorithm.

With the continuous evolution of the internet, internet open platforms have been constantly changing, resulting in diversified user groups and application scenarios, which in turn have led to various service models. To promote the development of internet open platforms, it is necessary to explore these new scenarios and service models. This paper aims to discuss the technical architecture, service models, and operation of internet open platforms. Initially, we review the development history of internet open platforms and introduce some typical internet open platforms. Additionally, we analyze the technical architecture of internet open platform, and introduce four different product forms and compare their advantages and applicable scenarios. Furthermore, we summarize five service models of internet open platforms, including the strategies adopted by enterprises at different stages. Finally, we analyze the developers of open platforms and provide suggestions for internet platform operation. We found that both enterprises and developers can fulfill their demands and achieve win-win cooperation through internet open platforms. Furthermore, the service model of open platforms will change with the development of enterprise products. Thus, companies should choose suitable service models based on their goals and market positioning. This study serves as a reference and guidance for the construction and operation of internet open platforms.

Matching supply and demand is a key link in the process of online trading information system design. In order to alleviate the problems of low efficiency of online trading , leakage of private information of buyers and sellers, inaccurate information and high cost of platform supervision, a blockchain-based matching trading model is proposed and designed based on the characteristics of multi-attribute matching trading, transferring the trading process to blockchain, making the trading decentralized, trustworthy and non-tampering, and realizing accurate matching of supply and demand between buyers and sellers based on ant colony algorithm, meeting the needs of both sides trading in the trading platform in terms of cost, efficiency and security. The experimental analysis shows that the model is compared with the traditional matching of supply and demand has timeliness, safety and effectiveness.

With the acceleration of global economic integration, cross-border transactions are becoming more and more frequent. However, traditional cross-border transaction systems generally suffer from inefficiency, high cost and privacy leakage. In this paper, we study the privacy protection problem in cross-border transactions and propose a privacy protection scheme for cross-border transactions based on block-chain technology. First, this paper briefly introduces the basic principles of block-chain technology and its application scenarios in cross-border transactions. Second, to address the shortcomings of existing privacy protection technologies, this paper proposes a privacy protection scheme that combines data encryption technology, zero-knowledge proof and hybrid network. The scheme achieves automatic execution and verification through smart contracts, which effectively protects the privacy of cross-border transaction participants. Finally, the effectiveness of the privacy protection scheme proposed in this paper in improving the level of privacy protection for cross- border transactions is verified through experimental evaluation and result analysis. This study is important for promoting the development of cross- border transactions, improving transaction efficiency and protecting user privacy.

In power grid, A fast and effective solution is needed to improve the efficiency of IP phone activation, migration, and troubleshooting. This solution requires a simple activation and migration process, using head and tail ends as a PLC relay system, ultimately achieving the goal of IP phone access to the network without wiring. In addition, a filtering algorithm was designed to adapt to specific field requirements in the power grid. Test results show that under a 100MHz power line carrier, the transmission rate can reach 34.8MBps, fully meeting the access requirements of IP phones in the power grid. This provides an effective way to improve the efficiency of phone activation, migration, and troubleshooting.

Currently, quantized neural networks have been widely applied in edge device model inference. In model inference using FPGAs, convolution operations are typically implemented using DSP that can provide 27×18bits operations. However, there is no general method for fully utilizing DSP to compute arbitrary low-precision quantized data in parallel. In the paper, we propose MDP, a universal solution that packs multiple multiplications in one DSP based on quantization precision and convolution parameters in multiple modes. To maximize the DSP multiplication performance, we design a data pipeline architecture based on GEMM. The experimental results show that, for single convolutional layer, MDP can decrease DSP resource utilization by 0.78× while maintaining the same level of parallelism. Compared to UltraNet, MDP achieves 1.56× latency improvement.

Structured multi-classification is a typical supervised learning problem, where each training instance is equipped with a set of candidate labels among which only one is the true label. This problem is often plagued by the slow training speed and the low classification accuracy. The goal of this paper is to propose a novel framework called normalizing quadratic Mapping and principal component Dimensionality Reduction analysis (MDR) Neural Network. Thorough experiments demonstrate it sets the new state of the art, accelerating the convergence of neural network and improving the accuracy of classification recognition with a few number of training rounds.

With the continuous development and progress of information technology and image sensors, the amount of data contained in digital images is increased. Therefore, the image feature extraction and downstream applications have begun to receive widespread attention by computer vision field. However, the majority of existing contributions are concentrated on the utilization of machine learning models and ignore the graph convolution theory can also be utilized to extract the small differences features from input images. In this paper, we utilize the graph convolution model to dispose the issue that small differences between classes make sample differentiation difficult in the process of feature extraction with optimization loss functions in the feature space. From our experimental results, we can conclude that our proposed model can achieve the image feature extraction and validate the identification accuracy reaching to approximately 95% for image classification task with reasonable system costs.

In this paper, big data of Gen Z new media by relevant technology is collected and analyzed, and the structure of Gen Z employee image is structured and described, making group characteristics of enterprise employees and the precision information push possible. This study efficiently collects, integrates, and analyzes data, and more accurately reflects users’ preferences to meet the needs of today's growing Gen Z market with Beautiful Soup, as well as Python and other algorithms. We calculate the optimal number of clusters with the K-prototype clustering algorithm through the silhouette coefficient based on big data. The results provide a solution for tracking the user's dynamic data and accurate preferences and interests.

Traffic signs contain important traffic information. The traditional traffic sign detection method cannot solve the problem of low detection accuracy caused by the small, occupied area of traffic sign images. Based on this, a traffic sign detection algorithm based on improved YOLOv4 is proposed. Firstly, the 13 × 13 large receptive field detection layer is removed on the YOLOv4 structure, and the 104 × 104 detection layer is added. It obtains more global feature information and improves detection accuracy. The attention mechanism is introduced into the algorithm, that is, the backbone network extracts three feature layers and then adds the scSE module. Make the network focus on the target area and improve the algorithm detection ability. Secondly, in order to speed up the convergence of the network and improve the detection accuracy, a dynamic residual connection is added to the backbone network. It promotes the spread of well-performing signals. And use the decoupled-head detection head to use different branches to calculate classification and positioning tasks. By evaluating the average accuracy of the detection effect of CCTSDB traffic sign data set, mAP reaches 97.68 %, which is 3.78 % higher than YOLOv4. Moreover, network convergence experiments have shown that the improved model converges faster. Compared with other models, the improved model has better detection performance for smaller traffic signs and can better meet the actual needs of high-precision detection.

Attribute-Based Asymmetric Group Key Agreement (AB-AGKA) protocol allows group members who meet attribute requirements to agreement a common secret, obtain a symmetric session key from it, and implement fine-grained access control. However, since the authentication of group members in the AB-AGKA protocol is based on user attributes, once users disclose their information to others, the system cannot identify the traitor. Chameleon hash is a special hash function with trapdoors, and only those who have trapdoors can calculate hash collisions. Therefore, this paper combines an attribute-based signature scheme and chameleon hash algorithm to design an Attribute-Based Chameleon Signature (AB-CS) scheme, which addresses the tracking problem of dishonest group members during protocol agreement. Finally, an efficiency analysis of AB-CS is conducted, and the results demonstrate that our scheme is efficient.

The utilization and advancement of Internet of Things (IoT) gadgets have noticeably increased in recent times. Various sectors, including intelligent homes, healthcare, sports analysis, and different industries, use IoT devices. Traffic is a crucial aspect of IoT devices, and it differs from conventional device traffic in several ways. This investigation employed 41 Internet of Things (IoT) devices. A machine learning model was developed that can classify IoT devices into multiple categories based on 13 different network traffic features. The raw data that was gathered was preprocessed using techniques like standardization and scaling techniques of the dataset. Algorithms for feature selecting can be utilized to extract characteristics from textual data. Following stratification, the dataset contains 106,234 sets of features that were utilized to enhance the predictive model. This research employed various evaluation metrics to demonstrate the effectiveness of the logistic-boosted algorithms. In this research, collaborative machine learning methods were utilized as a starting point to develop an Intrusion Detection System (IDS) that can identify abnormal network traffic. The principal aim of this investigation is to identify attacks and abnormalities within a Smart Home IoT setting. We have introduced a new technique for constructing Logit-Boosted algorithm, namely Logi-CatBoost Classifier. Once the logit-boosted algorithm was employed to classify the dataset, Logi-CBC performed the best, with an accuracy of 88.70%. We suggested Logi-CBC algorithm achieved the top precision among previously used Logit-Boosted algorithms in past studies on the same dataset.

This paper proposes a delay optimization scheme for railway time synchronization networks based on delay jitter to address the problem of decreased synchronization accuracy caused by path delay jitter in the time synchronization process. Firstly, the impact of path delay jitter on railway time synchronization networks is analyzed, and the synchronization error is quantitatively analyzed. Then, a corresponding delay optimization scheme is proposed to address the error. The scheme first achieves basic clock compensation for the client clock, and then obtains the optimal estimation of client clock compensation through an extended delay measurement mechanism implemented within the same period. This enables secondary compensation for the client clock, reducing the impact of path delay jitter on synchronization accuracy. Finally, a network time protocol simulation model is built in OPNET software to validate the proposed delay optimization method for railway time synchronization networks. Experimental results demonstrate that the proposed method improves the synchronization accuracy and stability of clocks in railway time synchronization networks.

With the expanding market demand for new energy power batteries, the industrial requirement for detecting laser welding defects of lithium battery’s pole are also increasing. Since the traditional machine vision defect detection has the problems of low detection accuracy and slow speed, this paper proposes a laser welding defect detection model based on the improved YOLOv5 algorithm to meet the requirements of industrial production. First, all C3 modules of the backbone was substituted for the newly proposed PConv module, which not only reduces the amount of redundant computation, improves the detection speed, but also increases the feature extraction efficiency of the backbone network for input images; second, we add the attention mechanism SE to the penultimate layer of the backbone network, which enhances the useful features of the backbone network for input images without additionally increasing the computation. This approach enhances the backbone network's ability to extract useful features from the input images and improves the accuracy of the model. The experimental results show that the improved model improves the average detection accuracy by 1.1% to 97% compared with the original YOLOv5 model. The FPS reaches 333, which can meet the current industrial demand for real-time defect detection.

Traffic flow prediction is a key technology in intelligent transportation systems and plays a crucial role in intelligent transportation. How to obtain dynamically changing spatial relationships of road networks without prior knowledge is a challenge in current traffic flow prediction. In response to the spatiotemporal dynamic changes of traffic flow, this paper proposes a traffic flow prediction method ASTCNN based on attention and time convolutional networks. This method utilizes time convolution to extract long-term dependencies in traffic flow time series; Based on an adaptive node embedding attention mechanism, model the dynamic correlation of different road network nodes without requiring prior knowledge of the graph; At the same time, by stacking time convolution and attention network layers, dynamic spatial correlation and nonlinear time dependence at different time levels can be effectively excavated, enhancing the representation ability of the model's spatiotemporal modeling. The experimental results on two real traffic datasets have shown that this method is concise and effective, and its predictive performance is superior to common baseline models.

This paper focuses on the research of the evaluation model of information communication and digitization capability in the new power system. Through the analysis of information communication network and digital technology in the power system, a method for evaluating the information communication and digitization capability of the power system is proposed, and the effectiveness of the proposed method is verified through experiments. Specifically, this paper first introduces the current situation and trend of information communication and digitization technology in the power system, and then elaborates on the importance and difficulties of evaluating the information communication and digitization capability of the power system and proposes an evaluation model based on the fuzzy hierarchy analysis method and entropy weighting method. Next, this paper details the design and implementation process of the model, including the construction of the index system, the determination of index weights, data collection, and processing methods. Finally, this paper verifies the effectiveness and feasibility of the proposed evaluation model through experiments. The results show that the model can accurately evaluate the information communication and digitization capability of the power system, providing a scientific basis for the construction and management of the power system.

With the development of web technology, web pages become more and more complex. Using the virtual DOM method can make the web page have efficient update performance without reducing development efficiency. An essential part of this method is the ability to quickly calculate the differences between the old and new virtual DOMs and update them into the old DOM. This paper presents a multi-threaded virtual DOM difference method based on web workers. This method improves the performance of difference method by offloading tasks to worker threads and using a multi-threaded parallel method based on worker thread pools, and presents a method based on SharedArrayBuffer to reduce the consumption of data communication between threads. Experimental results show that compared with the original main thread difference method, the new method can obtain a speedup ratio of 1.3x-2.4x in the scenario of 10k scale nodes and four threads.

Graph Neural Networks (GNNs) can better handle non-euclidean space data, and when combined with Federated Learning (FL) can address their data volume shortage and privacy security issues during model training. However, the traditional centralized Graph Federated Learning architecture faces threats such as unreliable centralized servers and attacks from malicious clients, and there may be missing edge structures between subgraph data of different clients, which reduces the performance of GNN models. In this study, a blockchain platform is introduced to implement decentralized Graph Federated Learning to improve the robustness of models. An approximation evaluation mechanism for finding similar subgraphs through Transfer Learning (TL) is also proposed, which can discover subgraph clients with structural associations and aggregate models of similar subgraphs without leaking data, thus solving the structural missing problem. Simulation results show that compared with existing algorithms, the approximation evaluation mechanism in this paper can better compensate for the missing subgraph structure and significantly improve the accuracy of model prediction, and the introduction of blockchain also effectively prevents the influence of malicious nodes and ensures the stability of the training process.

To reduce the dependence of lesion detection algorithms on annotated data and fully leverage the advantages of medical image big data from healthy people, this paper proposes an unsupervised breast lesion detection model based on a parallel convolutional neural network. It treats breast lesion detection in mammograms as a one-class classification problem. The network comprises two parallel convolutional neural network branches. One branch utilizes normal mammographic images as training samples to extract the deep features of normal images and considersthe compactness measure of features as the training loss. Simultaneously, to prevent the lack of inter-class discrimination in features extracted from the first branch's sample learning, the other branch introduces a subset of the ImageNet dataset as training samples and employs the descriptive measure of features as the training loss. This ensures that the extracted features not only satisfy intra-class similarity but also maintain inter-class discrimination. ResNet50 is chosen by comparison as the backbone network for the parallel branch and further improved. Experiments are conducted using the INbreast and BCDR datasets, and the results indicate that the proposed unsupervised lesion detection method could achieve comparable performance with the supervised methods without requiring any lesion annotation data.

The communication method in which both parties could modify certain types of shared resources without attracting the attention of other users is called covert communication. Generally, researchers believe that blockchain, as a distributed storage database that can still guarantee security in an unreliable network environment, has excellent characteristics such as open, decentralized, untrusted, anonymous, tamper prevention, traceability, and contract execution automation. More and more studies have applied it to covert communication technology. Due to the different transaction mechanisms of various blockchain systems, schemes based on system mechanisms are usually designed for a specific transaction mechanism of a blockchain, making the scheme more covert. The blockchain covert communication scheme based on transaction mechanism designed in this paper mainly utilizes the Bitcoin transaction mechanism in the Bitcoin system. The transaction parties generate addresses as input addresses through pre negotiated keys, construct invalid transactions through transaction withdrawal mechanisms, embed secret information in invalid transactions, and make them undetectable on the blockchain network. The receiver obtains information about invalid transactions through the negotiated keys, Decrypting and obtaining secret information completes the communication process of covert communication.

Traditional wireless sensor networks are vulnerable to data interception and tampering due to their low security. In order to meet the data security requirements under the current data explosion network environment, based on the traditional wireless sensor network, an intermediate node storage node is added between the sensing node and the base station to form a two-layer wireless sensor network, which makes the data management more secure, and makes the network expansion, deletion, and maintenance more convenient. At the same time, in the process of data encryption, lightweight national encryption algorithms SM2 and SM3 are used to improve the speed of data encryption while reducing the hardware demand for data. Therefore, a network model of "two-layer wireless sensor network data encryption based on national security algorithm" is proposed.

This article proposes a deep learning method for recognizing dance action videos. In this study, a dual path convolutional neural network is proposed to improve the traditional machine learning method. This algorithm can extract pixel information from video images, as well as the differential information of video actions. The Golden Tower algorithm is used to extract the changes in the lattice over time as the change features of convolutional neural networks. This study takes dance action recognition as an example and tests a given dataset. A comparison was made between the existing algorithm Inception V3 network and 3D-CNN network. The test results show that compared with the Inception V3 network, the algorithm proposed in this paper improves F1 by 10.90%, and improves recognition accuracy by 10.85% and 5.27%, respectively.

The line loss in low-voltage substation area is difficult to identify and the identification accuracy is low. In order to improve the accuracy of line loss judgment, a method based on multi-dimensional feature constrained clustering is proposed to identify sectional line loss anomalies in low-voltage substations. First of all, use LOF algorithm to collect electricity information, judge abnormal information of line loss data, and eliminate irrelevant data. Secondly, Hasusdorff distance formula is improved to calculate the multi-dimensional feature similarity of line loss data in power grid substation area, and the similarity matrix between line loss data is obtained. Finally, the multidimensional Hasusdorff distance is applied to the hierarchical clustering algorithm to identify the anomalies in the line loss data of power grid substation areas. The experimental results show that the proposed method has a good effect on identifying the line loss of low-voltage stations.

Imagined speech is a kind of intuitive paradigm of Brain Computer Interface (BCI). In recent years, the accuracy of the classification based on the electroencephalogram of imagined speech has increased owing to the more and more complex models. In order to further improve the accuracy and reduce the complexity of the model to a practical level, this paper proposed an efficient architecture consisted of Large Kernel and ConvMixer. This architecture increased the accuracy to 75.73%. The performance of the architecture is higher than other models, including combination EEGNet and Transformer which is the best model in the past. The results of further experiments show that Large Kernel can extract more task-related features from the EEG of imagined speech. ConvMixer can achieve a certain degree of performance improvement even with a greatly simplified structure of model.

Resource Description Framework (RDF), proposed by W3C, has become the de-facto standard for representing knowledge. Keyword query is a simple and effective way to analyze RDF data, which has been extensively studied. However, in many application scenarios, RDF may contain ambiguous information. The fuzzy RDF graph model is an important way to describe this fuzzy information. Unfortunately, keyword query methods for this fuzzy RDF graph model are rarely covered in the literature. In order to provide an effective tool for general users to analyze fuzzy RDF, this paper proposes a fuzzy RDF keyword query method. The main idea of this query method is to first retrieve the corresponding keyword elements in the graph by keywords, then obtain the shortest path between the keyword elements by graph search methods, and finally merge this shortest path into an answer tree. Experiments show that the method in this paper has higher query efficiency and better query quality compared to the baseline method.

Speech recognition is an Al-enhanced technology converting the human voice from analogue to digital form to be further processed. Speech recognition can be applied to multiple fields including banking, health care as well as all the commonly used activated digital assistants such as Siri, Alexa, and Cortana. Recently, with the development of deep learning, a higher level of speech recognition. In this research, I am going to explore deep learning's function in communication- or more specifically, in emotion recognition in the context of Alzheimer's diseases. The key to dealing with the problem is based on finding an objective method to normalize emotion, match each emotion with the correct label and test the model with unstudied data samples. Characterized sound data are processed by extracting MFCC value and then trained with a CNN-based model. The algorithm is then applied to new data to test performance. Models are evaluated by comparing the result of each of the ten emotions evolved. Our result shows that female-angry has the highest recognition accuracy while male-fearful has the lowest. Later, simple sentences on different topics will be added as a package that can be called flexibly according to the need of various users and also as a standby in a situation when detecting negative emotions. This paper provides a detailed explanation of the app developed for Alzheimer's patients which are characterized by its ability to recognize emotions and make corresponding reactions. I construct the architecture, examine the strengths and challenges of the model, report performance and test its function in real-world situations.

Aiming at the cold start problem and the limited prediction performance associated with sparse data in drug side effects frequencies study, we propose a matrix factorization method called Social Recommendation algorithm based on Multi Feature Fusion (SRMFF). First, the trust propagation mechanism is introduced into the matrix factorization, and the data sparsity and cold start drugs are alleviated by factorizing the drug side effect score matrix and drug trust matrix concurrently. Second, we select SIDER dataset and the drug Azvudine against COVID-19 as the experiment data. The results show that SRMFF can predict the frequencies of drug side effects effectively and have excellent performance in multiple performance indicators. At the same time, SRMFF can also predict the frequencies of side effects of new drug in the population.

The research on speech recognition shows that the accuracy of speech recognition in practical application scenarios is still not high because speech is affected by the user's unclear pronunciation, environmental noise and other factors. A key technology of speech recognition based on deep learning is proposed for resource browsing in campus enrollment. Firstly, the computational advantages of deep learning and the problems in the process of audio acquisition are analyzed. FFT algorithm is proposed to complete the noise reduction and enhancement of speech, and LSTM algorithm to complete the enhancement and recognition of speech. The experimental results show that FFT+LSTM algorithm has a high accuracy of speech recognition, reduces the response time of speech recognition, and has better flexibility and practicability.

The Non-Line-of-Sight (NLOS) environment significantly reduces the accuracy of Ultra-Wideband (UWB) ranging and positioning in the UWB positioning system based on time of arrival. The mainstream method is to identify the Line-of-Sight (LOS) and NLOS environment through machine learning and make corresponding corrections to improve the accuracy of the positioning system. However, the existing research and application of machine learning methods do not fully consider the situation that the test set exceeds the training set, which is inconsistent with the actual scenario. Firstly, in this paper, we show through experimental results that when the test set and the training set are independent, different training set acquisition methods will directly affect the accuracy of machine learning. The denser the training set, the higher the precision and the greater the workload. Therefore, aiming at how to choose the training set collection scheme, we put forward an evaluation index to determine the best training set collection scheme. Commonly used machine learning algorithms, such as Random Forest (RF), and XGBoost, are used here to explore the parameter configuration of this index. The research results show that when using machine learning method to identify ultra-wideband NLOS scenes, the acquisition step size of training set is 1.6 meters, which can make the identification accuracy high and the acquisition workload small.

With the trend of clean energy gradually replacing traditional energy as the main energy, the risk assessment of the electric energy spot market is facing new characteristics. Firstly, the paper analyzes the risk characteristics brought about by the two new features of the high-proportion clean energy spot market, the time-of-use price signal of electric energy is not clear, and the cross-regional electricity market and the multi-electricity commodity market are coupled together. And then a risk transmission chain for clean energy units is built to participate in the power spot market. Secondly, a risk assessment model based on CEEMDAN-LSTM for the high proportion of clean energy electricity spot market is constructed. Among them, the CEEMDAN method solves the problem that the risk factor sequence in the high proportion of clean energy electricity spot market is non-stationary and it is difficult to extract new features of the cycle. The LSTM model fully identifies the time-dependent characteristics of risk factors. Finally, the paper verifies the model based on the actual market data.

Sentiment Analysis (SA) is one of the most challenging tasks in natural language processing. It is the process of analyzing, processing, generalizing, and reasoning about subjective text with emotional overtones. To cope with sentiment analysis tasks, current deep learning approaches typically use large-scale pre-trained language models and attention mechanisms that apply computed attention weights without any restrictions on attention allocation. This mechanism is not an ideal model for dealing with the task of sentiment analysis, because in all kinds of language used in life, a semantically complete sentence is composed of words with different semantics connected to a complex and diverse grammar. Factors influencing the sentiment tendency of a sentence include the meaning of words and the change in semantics by the superposition of grammars with different structures. This paper proposes a sentiment analysis model that incorporates information about dependency syntax structure. Dependent syntactic information treats words as nodes and establishes syntactic grammatical structures between words by establishing node-to-node relationships. The undirected graph composed of the dependent syntactic structure information screens the text embedding derived from the attention mechanism in multiple layers, so that each layer focuses only on the associated text, and the sentiment classifier uses the text information containing the syntactic structure of the grammar in the classification. Experiments on the Online_Shopping_10_Cats dataset show a significant improvement in the final results compared to current state-of-the-art sentiment analysis models.

Cervical cancer is malignant tumors for women. It is crucial to delineate cervical cancer accurately from CT images. Existing tumor segmentation networks have problems in the cervical cancer segmentation process, such as insufficient feature extraction, poor universality, and low segmentation accuracy. We propose a Full-scale network combined with Cascading-attention and Dilated convolution (FCD- et). Firstly, we used transformer module to provide better global information extraction capabilities. Then, more detailed semantic information is obtained through the feature fusion module at the skip connection. Finally, we use dilated convolution and Squeeze-and-Extension network (SENet) to expand the receptive field, thereby extracting more characteristic information. Comparing the proposed method with mainstream methods, the experimental results show that our proposed method has significant advantages in various evaluation metrics and provide higher accuracy for cervical cancer segmentation.

Proton radiography has a broad application prospect in proton radiotherapy. However, it is often affected by scattered protons leading to lower image quality than x-ray images. Traditional image denoising methods can change the measured Water Equivalent Path Length (WEPL), resulting in severe radiation damage to patients. This paper proposes a new proton radiography denoising method. This method focuses on the imaging mechanism and puts forward the corresponding relationship between the distortion of WEPL characteristic curve and energy and share of scattered protons firstly. Then, the powerful feature extraction capabilities of deep learning networks are used to remove the noise to obtain high-precision proton images. Experiments show that this method improves the quality and effectively corrects the WEPL.

Understanding the distribution range of bamboo resources in Yiliang county is important for the ecological services and economic benefits of the county. In this paper, we investigated the distribution of bamboo resources by using 3S (GPS, RS, GIS) technology with remote sensing images and field survey. The survey results show that: the total area of bamboo forest in the county is 65863.72×10^-2 km² ; the origin of bamboo resources is mainly plantation and natural forest; the types of bamboo forest are pure bamboo forest, tree forest and mixed bamboo forest; the applications are mainly nature reserve bamboo forest, protective bamboo forest, timber bamboo forest and bamboo forest for bamboo shoots. There are nine kinds of bamboo species, including Jinfo Mountain Fang, Xiaochaoba Fang, Yushan Bamboo, Cixi Bamboo, Mao Bamboo, Bambusa Bamboo, Shui Bamboo, Yongsan Fang and Siban Bamboo. The protection level is Ⅰ, Ⅱ, Ⅲ and Ⅳ respectively.

The development of ceramic technology cannot be separated from traditional culture. China is a country with a long history and has left many excellent works in various dynasties. The development of ceramic technology has attracted more and more attention. Ceramic process product design needs more innovative ideas to meet market demand. Therefore, in this paper, machine learning technology is used to carry out data statistics and creative recommendations on product design, improve the design sense of ceramic technology and meet the needs of the masses. Qualitative and quantitative analysis and experimental methods are mainly used in this paper, with the help of machine learning, the related research on the creative design of ceramic craft culture. The experimental data show that in the field of ceramics, the training accuracy and test accuracy of the SVM model are kept above 80%, among which the weighted SVM algorithm has a higher grasp of the product accuracy.

To investigate the research hotspots and trends of hydrophone technology around the world, this paper visualizes the literature based on authors, countries, journals, and keywords since the twenty-first century using Citespace literature analysis software, which can help related workers better understand and grasp the research dynamics and hydrophone development trend. It has been discovered that research on hydrophones has been continuously rising, with China and the United States having the most papers in this sector. Various forms of hydrophones, including piezoelectric, MEMS, and fiber optic, offer promising development potential. Furthermore, the author feels using bionic technology for fiber optic hydrophones is reliable.

The rocky desertification, as the number one ecological problem facing the southwestern region of our country, is the result of the extreme degradation of land in the karst area. The Lingui District of Guangxi Guilin City is listed in the 13th Five-Year Plan for the rocky desertification project. In order to explore the sensitivity of rocky desertification ecological fragile area to the development of ecological environment, taking the rocky desertification phenomenon in Lingui area as an example, based on DEM image, 2020 Landsat8 remote sensing image and the third land survey data, ENVI software, Arcgis software and GIS technology are used to divide the areas with high ecological constraints, high ecological constraints, moderate ecological constraints and low ecological constraints. This will help to analyze the constraints and influences of ecological environmental vulnerability caused by rocky desertification on spatial development, and provide reference for the ecological environment control, development and ecological restoration of rocky desertification area in Guangxi.

Timely and accurate identification of peach tree pests and assistance in pest control is the key to promoting the healthy development of the peach industry. A lightweight peach tree pest image recognition model based on global context modeling is proposed for existing pest image recognition models with high structural complexity, high dependence on computational resources at runtime, and deployment difficulties. First, a lightweight global context modeling module is embedded in the ShuffleNet V2 unit to aggregate all the location information of the input image, obtain global context features, correct the pest information, and improve the feature representation capability of the model. Then, the ordinary convolution layer of ShuffleNet V2 is optimized to add lightweight convolution to improve the feature extraction ability of the model at different stages. Finally, a softmax classifier is connected to obtain a lightweight peach tree pest image recognition model. After experimental verification, the proposed model outperforms neural networks such as AlexNet in terms of recognition accuracy, parameter computation and floating point operation, and the number of parameters is only one-tenth or even a few tenths of AlexNet, which is a lightweight and efficient neural network model that can be easily deployed.

This work aims to study the modelling method of braiding preforms based on the rotary 3D braiding process. In the rotary 3D braiding process, the carrier is driven by the horn-gear through its alternating rotation, thus yarns interlacing to form the braiding preform. Encoding the horn-gear and the carrier to study the movement law of the carrier. The trajectory of the carrier and the height of the braiding node jointly determine the spatial topology structure of the braiding preform. The yarn trajectory is fitted and optimized using the cubic B-spline curve, Python script is written to drive CATIA to establish a solid model of the braiding preform and a fabric structure simulation interface is developed using PyQt5. Experiments on the braiding of irregular components are conducted to compare the actual braiding fabric with the corresponding model. The result indicates that the braiding algorithm could accurately record the coordinates of the trajectory points of the carrier; The fitted yarn trajectory conforms to the actual trajectory of the yarn; The simulated preform structure is consistent with the actual braiding preform. Therefore, it can be concluded that the modelling method is suitable for structural simulation of various cross-section braiding preforms.

The single-phase grounding of underground cables (feeder lines) in coal mines causes leakage, which poses great risks to safety production. A method for selecting faulty lines using long- and short-term memory neural network (LSTM) is proposed to address this issue. Firstly, a model of the coal mine underground power supply system is built, and a large number of simulation data are generated for training the neural network. Secondly, the LSTM neural network model is built and trained by using the keras platform. Finally, the trained LSTM neural network is tested using the new simulation data generated in various operating conditions. The results indicate that the LSTM neural network can quickly and effectively distinguish the faults of bus or line and select the faulty line correctly. This method is not affected by voltage phase angle, grounding resistance value, voltage level, number of feeder lines, and intermittent arc, thus proving its effectiveness.

Perceived value of knowledge-based payment plays a decisive role in consumer behavior and decision-making. Studying the "black box" effect mechanism of perceived value of knowledge-based payment on consumer purchasing intention can help optimize marketing strategies of knowledge payment platforms. In this paper, we propose a method for evaluating and predicting consumer satisfaction based on Convolutional Neural Networks (CNNs). Firstly, we collected and preprocessed a large amount of consumer review data and transformed it into a form that could be processed by our CNN model. Secondly, we designed a CNN model that combined convolutional layers, pooling layers, and fully connected layers, and trained it using backpropagation algorithm. Finally, we used this model to evaluate and predict new consumer review data. Experimental results showed that our method effectively identified sentiment information in reviews, accurately evaluated user satisfaction, and predicted future satisfaction trends with relatively high precision. This proposed method can provide a more reliable means of evaluating and predicting consumer satisfaction for businesses, and has the potential for widespread application in the commercial sector.

To address the low accuracy of TransUent network in small-area feature recognition, two feature classification networks based on improved TransUent network are proposed: DS-TransUnet network and DS-A-TransUnet network.DS-TransUnet network introduces a residual module in the bottom layer of the Convolutional Neural Network (CNN) module of the hybrid encoder of TransUent network to deepen the network and uses convolutional operation instead of maximum pooling in the hybrid encoder to reduce the loss of feature information. The DS-TransUnet network introduces a residual module at the bottom of the Convolutional Neural Networks (CNN) module of the TransUent network to deepen the network, and uses convolutional operations instead of maximum pooling in the hybrid encoder to reduce the loss of feature information, and uses depth-separable convolution instead of regular convolution to reduce the residual blocks and the large number of parameters caused by convolutional operations; the DS-A-TransUnet network On the basis of DS-TransUnet network, a void space pyramidal pooling module is introduced, and the classification accuracy is further improved by using void convolution with different expansion rates in parallel to obtain different size of perceptual fields and fully extract multi-scale features. The experimental results show that the recognition accuracy of DS-TransUnet network is 3.06% and 2.85% higher than that of TransUent network for arable land and water bodies, respectively, and the overall recognition accuracy reaches 89.14%; the recognition accuracy of DS-A-TransUnet network is 3.17% and 4.79% higher than that of TransUent network for arable land and water bodies, respectively, and the overall recognition accuracy reaches 90.53%. The overall recognition accuracy reached 90.53%.

The ultimate condition of fiber reinforced polymer (FRP)-confined noncircular concrete columns subject to axial compression is determined by the rupture of FRP jacket, which is estimated via the FRP strain efficiency factor (κε). A database including 189 experimental results for FRP-wrapped square and rectangular columns is established to evaluate the parameter κ_ε. The test data cover several types of FRP and different unconfined compressive strengths (f’_c0=21.1~110.8MPa). Basic statistical parameters, the influence of f’_c0, as well as the effect of nominal thickness on κε are discussed by means of statistical analysis. As a result, the relationships between the above influential parameters and κε are evaluated effectively.

According to the requirements of UN/WP.29 Rl56, the management system and vehicle technology requirements for vehicle manufacturers to use RX software identification number are put forward. In order to meet the compliance requirements of regulations, a set of RX software identification number management strategy has been developed to comprehensively cover the management cycle of RX software identification number from organizational structure to management process sett.ing to recording methods. It can not only guide OEMs and Tierls to identify, record and verify RX software identification number related information in the whole software upgrade process. Also, can help each enterprise to establish a perfect RXSWIN management mode.

Automatic sleep staging can provide an aid to experts and it’s an important basis for assessing sleep quality and diagnosing sleep disorders. However, the complex non-stationary and non-linear characteristics of Electroencephalogram (EEG) signal make it difficult to use them directly for sleep staging, while existing EEG signal decomposition methods have drawbacks such as modal mixing and boundary effects. To address these issues, this paper decomposes single-channel EEG signal into sub-bands signal based on a new decomposition method, Empirical Fourier Decomposition (EFD). The time-domain, frequency-domain, and non-linear features of these sub-bands signal are then extracted. A ReliefF algorithm is then used for feature selection to obtain the feature subset with the optimal staging performance. The feature subset is input to the Random Forest (RF) to class, and the staging results of the proposed method are also compared with existing literature. The accuracy rates of the proposed algorithm in 2-6 classes of sleep staging are 97.79%, 95.35%, 94.33%, 93.35%, and 92.72%, respectively. Due to the portability of single-channel devices, the proposed method can effectively facilitate the research and application of home sleep monitoring devices.

Large break loss of coolant accident is one of the benchmark accidents in nuclear reactor design, and its reflooding process involves complex two-phase flow and heat transfer, which has been a hot spot of research in various countries. Simulating the re-flooding process by thermal hydraulic analysis procedure is of great significance to ensure reactor safety and determine the design parameters of nuclear power plants. In this paper, the FLECHT SEASET experiment is selected to perform validation calculations on the reflooding model of RELAP5 program and cosSyst program, which proves the rationality of its calculation of re-flooding heat transfer. And the effect law of coolant inlet flow rate, system pressure, coolant temperature, initial temperature of the envelope, and heating rod power on the re-flooding process was investigated by cosSyst.

Dynamic network defense technology refers to the application of dynamic defense technology at the network level. It implements dynamic defense against harmful attacks on the target network to improve the difficulty for attackers to identify networks and penetrate network nodes. Most of them focus on traditional networks and ignore the advantages of software-defined networks (SDN). Network feature obfuscation is a successful reconnaissance countermeasure method, which will interfere with the information collected by the attacker in the reconnaissance stage and make the attacker's attack invalid. In this paper, a dynamic defense system is proposed to timely, dynamically and adaptively evaluate the effectiveness of SDN based deception defense system techniques. In the future, we will further study how to improve the dynamic defense capability of SDN, and analyze the development direction of dynamic defense based on SDN.

In recent years, with the rapid development of economy and urbanization, a large number of parks have been built all over the world. The number of fire accidents occurring in the parks is also increasing, resulting in tremendous casualties and property damage. These parks usually cover multiple sub-scenes, such as houses, vehicles and greenery, and each sub-scene has its own unique fire accident features. Existing fire detection methods cannot meet the diversity of subscenes in the park. To improve the detection capability of each sub-scene model, we propose a cloud edge collaboration framework based on transfer learning for training fire detection models with multiple sub-scenes in a park. The framework has two phases: initialization phase and update phase, using the rich data and powerful computing power in the cloud to assist the edge nodes with limited resources to train the models. First, the cloud integrates the fire data of all edge nodes for model training to obtain all fire features, and then each edge node uses its own data set to fine-tune the cloud-trained model after freezing some parameters. In this paper, we construct a fire image data set with multiple subscenes in a park and use the YOLOv5s model for experiments, and the experimental results show that the framework can improve the mAP of each edge node model. This paper provides a novel approach for fire detection, which is of positive significance for fire prevention and control.

This paper proposes an intelligent learning application based on OCR (Optical Character Recognition) text recognition and a personalized recommendation algorithm, which can help students to record their mistakes and recommend similar questions according to the type of mistakes so that students can further consolidate their mistakes and improve their correctness rate in their daily work. This paper is based on the JAVA language for development and design. The user side is presented in the form of a WeChat applet, which reduces the use threshold and increases the user's dependability. The project combines the functional features of OCR algorithms for regular and irregular text recognition. It adds the ability to recognize distorted and blurred fonts to print fonts, scanned text, and other horizontally positioned text, making it more relevant to everyday use scenarios. The personalized recommendation algorithm of the project adopts various recommendation algorithms respectively, among which the collaborative recommendation algorithm is divided into user-based collaborative Filtering and item-based multi-criteria collaborative filtering, which perform relevant recommendations for user positive feedback exercises and the collaborative recommendation algorithm divided into user-based collaborative filtering and item-based multi-criteria collaborative filtering, which can predict users' preferences more accurately. The project also incorporates an exercise recommendation method that models the user's knowledge state, which models the user's personalized knowledge state based on their daily problem-solving habits, and makes exercise recommendations based on the model, which is a good predictor.

In this paper, the author collects corresponding pictures with Chinese knots on the Internet and unifies their format as .jpg. After using Colabeler to label the Chinese knot in each picture, import the xml file and the xml file in pascal-voc format respectively. Put the prepared files into the corresponding folders in the yolo-master project to complete the production of the data set. After ensuring that the running environment of yolo-master is normal, adjust the corresponding parameters, run the train.py file and use this data set to train the model for detecting Chinese knots. The authors then test the model on a dataset of images mixed with other objects and obtain corresponding detection results. Finally, the authors analyze the results using metrics such as precision, recall, and F1 score. Furthermore, we discuss the novelty and future development potential of the method for object detection and recognition in detecting classical Chinese artworks.

Image style transfer is a technique for generating images with the help of computers. It can transfer the style features of image B to image A to generate image A with style B, which can not only meet the individual needs of users, but also meet the needs of users. The problem of insufficient image data for a certain style. In recent years, a series of models such as CycleGAN have achieved great success in the field of domain style transfer, but their training is unstable, and the problems of blurred, unstable, and inefficient generated images that are prone to mode collapse have not been solved, this paper proposes the GN-CycleGAN model for this problem. In this paper, a GN-CycleGAN model is proposed to solve this problem. By replacing the instance normalization in the discriminator with the gradient normalization, the discriminator space can be smoother without affecting the performance of the discriminator, so as to overcome the training instability of CycleGAN caused by steep gradient space. The experimental results show that, compared with CycleGAN, the loss function of GN-CycleGAN model can converge quickly under the same training time,, and it performs better on the monet2photo dataset. The generated images are in PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity) It is better than CycleGAN in two indicators.

An enhanced version of the YOLOv7-obb deep learning neural network was proposed to achieve fast and accurate recognition of eggplant fruits and stems. The improvements primarily revolve around two key aspects. Firstly, the inclusion of the CBAM attention mechanism enhances the network's ability to perceive and extract target feature information by expanding the perception field of the feature layer. Secondly, the integration of rich image information from the first and second layers of the network backbone enables effective fusion of multi-scale features. Additionally, a dedicated output layer in the network's neck is introduced to enhance the detection of small targets and improve the extraction of relevant feature information. To evaluate the model's performance, extensive training experiments were conducted using a carefully curated eggplant dataset. The results demonstrate a significant improvement in the mean Average Precision (mAP) of the new model, with an increase from 76.74% to 88.84%. This indicates that the enhanced model exhibits strong recognition capabilities, showcasing its potential for intelligent eggplant harvesting.

In order to improve the accuracy of hydropower forecast, a new power generation prediction method combining empirical mode decomposition (EMD), principal component analysis (PCA), sparrow search algorithm (SSA) and long and short term memory (LSTM) network model was proposed. Firstly, the original generated sequence was decomposed into multiple sequence components by adaptive noise global empirical mode decomposition to reduce the random fluctuation of the generated sequence. Then principal component analysis was used to extract the key influencing factors of feature sequences to eliminate the correlation and redundancy of original sequences. Finally, a linear LSTM model was constructed for each component, and the prediction results of the two models were combined and weighted to obtain the prediction results of each component. Aiming at the optimization problem of important parameters in the LSTM model, Sparrow search algorithm (SSA) is introduced to optimize the model parameters, sum and reconstruct the prediction results of each component, and get the final power prediction results. Taking the measured data of a hydropower station in Sichuan as an example, the experimental results show that the method is effective.

Global temperature change threatens our planet's ecosystems. This study introduces a new approach to address the problem of global temperature change. A global temperature prediction model was developed using LSTM and ARIMA techniques. The Pearson correlation model was used to investigate the relationship between temperature change and related factors, while an independent sample t-test model determined whether forest fires significantly affect temperature. The method has practical implications in climate monitoring and management, and high accuracy was achieved in the experiment. The findings provide valuable insights into the dynamics of global temperature change and lay the foundation for policymakers to implement evidence-based interventions to reduce the impacts of climate change on the Earth's ecosystems.

China's grassland area is vast, once a grassland fire occurs, it will cause serious threat to people's life and property. At present, the domestic grassland fire monitoring means is mainly manual monitoring, this monitoring means of monitoring cost is large, low efficiency, real-time poor, and it is difficult to achieve complete detection, so the use of satellite, UAV images and other advanced means to identify grassland fire is the current research hot spot. In this paper, using UAV images, the detection is divided into two stages according to the burning process of grassland fires. In the early stage, the improved YOLOv5 network model is used to identify smoke due to the more obvious smoke characteristics; in the middle and late stage, the color characteristics of the fire area change obviously, and a flame detection method combining static and dynamic features is proposed. The C3 structure of the Neck part of the YOLOv5 model is improved, and the CBAM attention mechanism is introduced to make the model pay more attention to the target information; the loss function of YOLOv5 is improved, and DIoU is adopted as its loss function, and the experimental results show that the mAP of the improved YOLOv5 model is improved by 2.53% when trained on the self-built smoke dataset, and the recognition accuracy is effectively improved. When flame recognition is performed, the flame is affected by wind, vegetation coverage, humidity and terrain, etc., and spreads in all directions at different speeds, forming a dynamic outward-expanding circle with an increasing area and a distinct black color in the overfire area. Static color features are firstly used for pre-determination, and then the dynamic area and contour changes of the overfire are further discriminated. Experiments show that the algorithm reduces the false recognition rate of grassland fires and has good practicality.

In the smart city development phase, the construction of community service facilities plays an important role in creating future communities and convenient living circles. With the continuous maturity of big data and GIS spatial analysis technologies, it is possible to make objective quantitative assessment of the construction of community service facilities and provide scientific decisions for future planning and layout. Our objective was to analyze the current spatial distribution characteristics of community service facilities in the Shangcheng District of Hangzhou, Zhejiang Province, China. We processed, counted and visualized the data from POI (Point of Interest) using the nearest-neighbor distance method and kernel density analysis in the ArcGIS platform. The results showed that, in terms of quantitative characteristics, the number of financial and cultural facilities was relatively low, making it difficult to achieve full coverage of residential land. In terms of layout, dining, transportation, shopping, and education facilities were arranged around the residential areas, while cultural, leisure and medical facilities were arranged along the city's primary roads or metro lines. In terms of clustering, the HB (Hubin) section was characterized by high density and large area clusters, indicating a highly developed system of community services and facilities. The GBNT (Genbei New Town) section showed a low-density and point pattern, indicating a poorly backward system of community services and facilities and needs to be improved and optimized.

The identification of volcanic pathways plays an important role in the study of volcanic reservoir formation and hydrocarbon migration. The volcanic channel of fissure eruption is different from the volcanic channel of central eruption, which often has the same reflection characteristics as the fault on the seismic section. The existing identification methods for the volcanic channel of central eruption cannot identify the volcanic channel of fissure eruption. In this paper, aiming at the accurate characterization of fissure eruption volcanic channel, a series technical process based on "enhanced processing attribute extraction attribute fusion" of post stack seismic data is established: after Principal Component Analysis (PCA) denoising of post stack seismic data, texture attributes are extracted to highlight the change of seismic reflection amplitude of fissure eruption volcanic rock, On this basis, local variance calculation is carried out to highlight the continuous difference between volcanic rock mass and surrounding rock, and the result is superimposed with the maximum likelihood attribute of identifying volcanic channel, so as to realize the accurate identification of fissured volcanic mechanism and channel, and reduce the ambiguity and uncertainty. This method has been applied to Yingcheng Formation in Songnan area.

Cardiovascular disease is a common disease that poses a serious threat to humanity, with a high incidence, disability, and mortality rate that is daunting. Carotid Atherosclerosis (CAS) is closely related to cardiovascular disease, so it is very important to accurately identify carotid plaque in ultrasound images for the diagnosis of stroke risk patients. At present, traditional ultrasound testing has strong subjectivity, poor reproducibility, and there are cases of missed detections. To address these issues, this article proposes an improved carotid artery ultrasound plaque recognition and localization algorithm. Firstly, preprocess the ultrasound image to reduce the impact of noise and shadows; Secondly, by improving the YOLOV5 model, setting custom anchor frames, and adding attention mechanisms, the final accuracy achieved was 3.48% higher than the original YOLOV5. The experimental results showed that the improved YOLOV5 model improved the recognition performance of the experimental dataset and effectively improved the network's ability to recognize the intima. Even in the case of small intra vascular differences and unclear neck boundaries, this provides new ideas for the research and application of medical image recognition.

The traditional port target recognition algorithm has the problems of high time complexity and low recognition accuracy. In order to solve this problem, we propose a method of Guangdong-Hong Kong-Macao Greater Bay Area port target recognition based on deep residual neural network. By studying various network models with different deep and parameters, a new multi-strapdown residual network model (Mu-ResNet) with less memory occupation and lower time complexity is designed by combining two-layer residual learning module with three-layer residual learning module. The experimental results show that the recognition rate of our proposed method is better than the traditional recognition algorithm on the port target dataset.

An asphalt road pothole image stitching algorithm is proposed to address the issues of low feature point matching accuracy, loss of pothole disease information, ghosting and deformation of the pothole when using traditional image stitching algorithms. The goal is to stitch the images captured by the pavement image acquisition equipment to form a complete asphalt pavement picture with the disease information of the pothole area intact. The stitched image's pixel point coordinates are transformed to the coordinates of the reference image using a single-strain matrix after first extracting the pit regions in the image and extracting feature points for these regions using the MSER-SIFT feature extraction algorithm. Next, the best matching point pairs are found using an improved random sampling consistency algorithm. Finally, a region-weighted averaging approach is utilized to merge the images. The experimental asphalt road images include pavement images with sufficient light during the day, pavement images with complete pothole disease information, pavement images with incomplete potholes, and pavement pothole images at night to test the effectiveness of pavement stitching under various conditions. The research results show that this splicing method not only improves the matching accuracy and speed of the traditional splicing algorithm, but also ensures the integrity and invariance of the pavement pothole information.

Due to the harsh operating environment of gas turbines, their performance status will degrade with the growth of service life, which can even lead to unscheduled downtime in severe cases. The purpose of building the model is to provide guidance for the operation optimization and maintenance decision of the unit, and to improve the safety and reliability of equipment operation as much as possible. This paper uses BP neural network algorithm to establish a gas turbine health status assessment model, and verifies the accuracy of the model assessment through example analysis and research. The findings indicate that the model has demonstrated good evaluation performance and holds a certain level of practical significance in gas turbine health status assessment.

When observing space debris with ground-based telescopes, in some cases relative motion occurs between the star and the telescope, creating a trailing tail. Motion-blurred star images are characterized by low signal-to-noise ratio and heterogeneous motion blur. In order to improve the recovery quality of motion blurred star images, this paper constructs a motion blurred star image dataset using simulated star images. The Residual Block in the MIMO-UNet is replaced by Res FFT-Conv Block to extract the spatial domain and frequency domain information of the blurred star image. Finally, training is performed on the improved MIMO-UNet. The experimental results show that the network can improve the energy of the trailing stars and recover the shape of the trailing stars.

At present, the image features extraction and classification have widely utilized to detect objects in numerous applications. Additionally, existing classification are primary concentrated on the machine learning models or train a convolutional neural network, which can obtain the acceptable extraction and classification accuracy. However, these methods ignore the graph convolutional operation can also be utilized to achieve the image features and classification tasks. In this work, we initially extend the graph convolutional neural network to the image features extraction and classifications, which can also obtain an acceptable classification accuracy with approximately to 95%. Initially, we utilize the clustering algorithm to mix multiple image pixels to a graph node and the adjacent matrix is depended on the pixel value features. Subsequently, the transferred graph is trained by a graph convolutional neural network to identify the class of these images. From our extensive experimental and compare with traditional machine learning models, our proposed model can successfully identify the subjects in the images with highest classification accuracy and reasonable computation costs.

In response to the high cost of obtaining high-quality annotated data, some methods have emerged that automatically extract annotated data from large-scale unlabeled datasets. However, these methods often result in noisy data, which can negatively impact the performance of models. Therefore, noise-handling methods for data have high practical value. This article proposes a novel noise-handling method based on textual entailment model and applies it to text classification tasks. Experimental results demonstrate that the proposed method performs well when the proportion of noise data in the overall dataset is relatively high, which provides a new approach for the research of noise-handling methods.

Single-Frame Infrared Small Target (SIRST) detection aims to extract small and dim objects from the background of heavy noise and clutter. Traditional detection methods rely on handcrafted feature templates and prior knowledge, which exhibit inferior robustness in complex environments. Despite achieving promising outcomes, the existing detection methods based on CNN still suffer from the issues of small targets being lost in deep layers and long-range dependency problems. In this paper, we propose a cnn-based enhanced coordinate attention network with centralized visual center (CVC-ECANet) for infrared small target detection. Specifically, features of small target can be extracted properly by densely nested structure, where the enhanced coordinate attention module (ECAM) in nodes can enhance the features and maintain them in deep layers. In addition, by incorporating the centralized visual center module (CVCM) after the deepest layer of the network, global long-range dependencies can be captured and shallow features can be centrally modulated. Extensive experiments were conducted on the SIRST dataset, and the results demonstrate that the proposed method performs well in terms of probability of detection (Pd), false-alarm rate (Fa), and intersection of union (IoU).

By utilizing computer graphics and image processing technology, we can effectively enhance the effectiveness of visual communication design. Specifically, using the principle of three primary colors to create a set of vector functions can more accurately calculate the partition model for visual communication. Firstly, we need to comprehensively consider all the characteristics related to the image, and then encode them in an orderly manner to meet our requirements. Secondly, the appropriate hyperplane is selected according to these characteristics to adjust the visual transmission mode. Ultimately, some creative image processing techniques are needed to improve their quality. After systematic research, it has been found that using visual communication design methods can achieve excellent results. It has strong stability, flexibility, and can exert the best communication ability in four types of complex images, greatly improving the level of precision in communication. In addition, this method further clarifies the application direction and methods of image processing technology, in order to provide theoretical reference for optimizing visual communication application solutions.

Existing object detection methods primarily focus on detecting large targets in images, with limited research conducted on small targets. Additionally, there are challenges with low detection accuracy and difficulty in meeting real-time requirements. Based on this, an improved RetinaNet is proposed. First, the use of high-resolution feature maps can better capture the features of small targets. Then a weight factor is added to solve the unbalanced distribution of training samples caused by using high-resolution feature maps. Finally, adding a novel query subnet by two main steps: first estimate the coarse location of the small target in the high-level feature map, then the exact location is calculated in the low-level feature map. Experiments show that this query subnet saves the computational cost of low-level feature maps, and greatly improved detection speed. The performance of the model is evaluated on a new benchmark called TinyPerson, mAP up to 36.18, FPS up to 13.98. Compared to the unimproved RetinaNet, the accuracy and speed are improved by 0.9 and 1.28 respectively.

With the rise of social networks and online communities, group activities become increasingly common, which promotes the development of group recommendations. As the first step of group recommendation, group discovery aims to identify potential groups from users. Current group discovery methods often fail to consider behavior preferences and social connections simultaneously and may not be trained in a group-discovery-oriented manner. To address the above problems, we propose a novel group discovery method called BSGD based on users' behaviors and socialization, which introduces three tasks and implements group-discovery-oriented end-to-end training through multi-task joint training. The experiment results show that the proposed method has an excellent performance in both group discovery and gains in group recommendation on two real-world datasets.