This PDF file contains the front matter associated with SPIE Proceedings Volume 12969, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

In this paper, a 3D dangerous goods detection method based on RetinaNet is proposed. This method uses the bidirectional feature pyramid network structure of RetinaNet to extract multi-scale features from point cloud data and trains the system using Focal Loss function to achieve fast and accurate detection of dangerous goods. In addition, in order to improve the detection accuracy, this paper also introduces the 3D region proposal network (3D RPN) and nonmaximum suppression (NMS) algorithm. The experimental results show that the proposed method performs well on our self-built CT dataset, with high accuracy and low false positive rate, and is suitable for dangerous goods detection tasks in practical scenarios.

The conventional automatic identification algorithm of double-feature fault signal waveform of power equipment mainly uses ART (Adaptive Resonnance Theory) network for classification and discrimination, which is easily influenced by the identification mapping relationship, resulting in low correct identification rate of fault signal waveform. Therefore, it is necessary to design a brand-new automatic identification algorithm of double-feature fault signal waveform of power equipment. That is to say, the waveform characteristics of dual-feature fault signal of power equipment are extracted, and the optimization algorithm for automatic identification of dual-feature fault signal waveform of power equipment is generated, so that the automatic identification of fault signal waveform is realized. The experimental results show that the designed double-feature fault signal waveform automatic identification algorithm for power equipment has a high correct fault identification rate, which proves that the designed double-feature fault signal waveform automatic identification algorithm for power equipment has good identification effect, reliability and certain application value, and has made certain contributions to improving the operation safety of power equipment.

Continuous updating and maintenance of feasible solutions is crucial when solving constrained multi-objective optimization problems (CMOPs). However, most existing constrained multi-objective evolutionary algorithms (CMOEAs) are not efficient enough in updating and preserving competitive feasible solutions, thus reducing population diversity. To address this issue, this paper proposes a dual-population (i.e., mainPop and auxPop) constrained multi-objective evolutionary algorithm with a feasible archive set for CMOPs, named DPFAS. The two populations have different functions in the algorithm. Specifically, the ݉ܽ݅݊ܲmainPop considers both objectives and constraints for solving the original CMOPs, while the ܽauxPop is used only for the optimization of objectives without considering constraints. In addition, a feasible archive set is used to store feasible solutions that are competitive in the ܽauxPop and provide useful information for the ݉ܽ݅݊ܲmainPop. Moreover, a fitness assignment strategy is designed to speed up the algorithm’s convergence. Particularly, the population converges faster by selecting better-nondominated solutions into the matching pool. Finally, experimental studies on 23 benchmark functions show that the proposed algorithm was more competitive compared with five state-of-the-art CMOEAs.

In response to the problems of low detection accuracy and inaccurate positioning in traditional YOLO algorithms for fruit detection tasks, this paper proposes an improved method - replacing the original loss function of YOLOV7 with the AlphaIoU loss function, and optimizing the boundary box regression of the model. This can accelerate the convergence speed of the model and improve training accuracy. Through experimental verification, it was found that the improved YOLOV7-tiny model significantly improved the average accuracy, accuracy, and recall in fruit detection tasks compared to the original network. This indicates that this method has good performance in solving the problems of traditional YOLO algorithm, providing better performance and accuracy for fruit detection tasks, and contributing to the further development of applications and research in related fields.

High resolution synthetic aperture radar system will produce aliasing problem in the sliding spotlight mode, we propose a solution based on sub-apertures splice to solving the program, the simulation and analysis of airborne mode are given. The method use the sub-aperture division, respectively, each aperture frequency shift, scaling, compression distance, azimuth compression performed prior to stitching, the data will be spliced after azimuth compression and windowing. Simulation results show that the algorithm can effectively solve the problem.

The brocade weaving craft has a long history, with exquisite patterns and profound cultural connotations. It is an excellent representative of Chinese silk culture and an eye-catching business card in the intangible cultural heritage of mankind. The process of making brocade is a very complicated craft. In order to be able to detect defects in time during the production process, an improved SE-SSD fabric defect detection algorithm is proposed for the low efficiency of defect detection in traditional production, the large model affects the deployment and the shortcomings of DB-YOLOv3. By improving the network structure and optimizing the prior frame adjustment mechanism, the algorithm improves the ability of model feature extraction and greatly reduces the parameters and calculation of the network. The experimental results show that the SE-SSD algorithm effectively improves the missed detection of linear and weak target defects. Compared with the SSD network, the detection accuracy is increased by 27.55%, reaching 93.08% mAP, the detection speed is increased to 49FPS, and the network parameters are reduced. 51.5%, which improves the practicability of the algorithm, and the ability to detect small target defects still needs to be improved.

Video facial micro expression recognition is difficult to extract features due to its short duration and small action amplitude. In order to better combine temporal and spatial information of video, the whole model is divided into local attention module, global attention module and temporal module. First, the local attention module intercepts the key areas and sends them to the network with channel attention after processing; Then the global attention module sends the data into the network with spatial attention after random erasure avoiding key areas; Finally, the temporal module sends the micro expression occurrence frame to the network with temporal shift module and spatial attention after processing; Finally, the classification results are obtained through three full connection layers after feature fusion. The experiment is tested based on CASMEⅡ dataset,After five-fold Cross Validation, the average accuracy rate is 76.15, the unweighted F1 value is 0.691.Compared with the mainstream algorithm, this method has improvement.

In order to enhance the efficiency and accuracy of homologous tampering detection, image segmentation algorithms and image feature points are combined. The Simple Linear Iterative Cluster (SLIC) algorithm is employed for image segmentation. However, manually presetting the number of patches is not applicable to all images and can influence subsequent segmentation results. To achieve a more accurate detection of tampered areas, this paper proposes a self adaptive image tampering detection algorithm. The number of image segments is determined based on image complexity, which allows the image to be segmented into semantically independent patches. Subsequently, the SIFT algorithm is employed to extract feature points for matching. Test results demonstrate that the proposed algorithm accurately localizes tampered regions and reduces algorithmic complexity.

Under the influence of high density operation and natural environment, the rail surface will appear abrasion damage, which will affect the safety and comfort of the train. Rail surface defect detection is an important part to ensure the safe and efficient operation of railway system. In order to distinguish whether there are defects on the rail surface, a method of rail surface defect image segmentation based on FPSO 2D-Otsu algorithm is proposed. The rail image is denoised and enhanced by adaptive fractional calculus, and then the rail image is segmented by FPSO 2D-Otsu algorithm. In order to verify the accuracy of the algorithm, the proposed algorithm is compared with PSO 2D-Otsu image segmentation algorithm. The experimental results show that the accuracy of FPSO 2D-Otsu algorithm in rail image segmentation is improved from 48.76% to 83.59% compared with PSO 2D-Otsu algorithm.

Looking at the traditional interior space design industry, the traditional design method is mainly manual design and the use of interactive modeling software and its design process mainly relies on trial and error. This paper takes the interior space design software platform as the background to study the collocation recommendation algorithm of the 3D home model, aim at improve the efficiency of the intelligent design algorithm. The recommendation idea of collaborative filtering is simple to implement, does not need to consider the inherent attribute characteristics of three-dimensional home projects, and is fast to calculate. After constructing the image feature database, this article uses the similarity between images to measure the visual similarity of the indoor space model; uses similar home projects to predict the collocation data of adjacent projects, and densifies the sparse collocation data; constructs each image separately Feature database, and use this to build its similarity table. According to the similarity table corresponding to each item, the first simNum items of the same category that are similar to the current item can be found. The experimental results show that compared with the traditional algorithm, the algorithm in this paper has greatly improved the accuracy of collocation recommendation.

The rule engine is an important part of the industry-education integrated Internet of Things teaching platform, and it is the basis for realizing the dynamic configuration of business rules in the practical teaching function. Combined with the data characteristics of the Internet of Things application scenario, this paper proposes a rule engine optimization algorithm based on Rete, and designs a pre-sorting algorithm based on rule frequency, which pre-sorts the order of nodes according to the frequency of use of rule patterns, with priority Match frequently used patterns, increases the sharing rate of nodes, and reduce the memory usage of the inference network. Through experimental simulation, the improved algorithm is verified, and the experimental results prove the effectiveness of the algorithm.

Aiming at the target distribution problem of anti-aircraft weapon firepower, a target allocation method based on multitarget particle swarm is proposed. The mathematical model of incoming target allocation constraint optimization is established, and the multi-target particle swarm algorithm is used to solve the target allocation model. The inertia weights in the velocity update formula of the particle swarm algorithm and the learning factor assignment method are improved. Compared with the simulation results and actual experience judgment, the designed algorithm solves the target problem in the air defense weapon system to a certain extent.

In view of the problems of excessive parameter setting and large calculation of YOLOv7 in pedestrian object detection in complex street scenarios, this paper proposes a lightweight method to improve YOLOv7 algorithm. Under the YOLOv7 framework, Partial Convolution (PConv) is integrated into the convolution of the original algorithm, replacing part of the convolution in the original convolution layer, and the SEAttention attention module is introduced to ensure the detection accuracy of the lightweight algorithm. The experimental results on the home-made data set show that, compared with the original YOLOv7 algorithm, the number of model parameters decreased by 11.0% in the improved YOLOv7 algorithm, and the algorithm calculation volume decreased by 19.4%, while ensuring the high accuracy of the original YOLOv7 algorithm. In this paper, the algorithm reduces the number of parameters and calculations, and achieves the balance of lightweight and accuracy.

In the course of work, people are easily affected by the complexity of the environment, so it is difficult to ensure the assessment of personnel safety risks. In view of this situation, a safety risk assessment method of key personnel in infrastructure projects based on image data coupling identification is proposed. Under the coupling recognition of image data, a comprehensive evaluation model is established by combining dynamic graph reasoning technology, and the relevant signals of the working environment of key personnel in infrastructure projects are extracted from the model, and images are obtained on the projection surfaces of the left terminal and the right terminal, so as to locate the positions of key personnel in infrastructure projects and realize the safety risk assessment of key personnel in infrastructure projects. The experimental results show that the average channel collision rate of the proposed method is low, which can greatly improve the accuracy of dangerous action identification and has good performance and effect.

DR (Diabetic retinopathy) a chronic progressive disease which affects eyesight and even causes blindness. It is significance to carry out the identification and severity diagnosis of DR, timely diagnosis and treatment of DR Patients, improve the people’s quality, especially the elderly, and improve the efficiency of diagnosis. In this study, with the goal of efficient and accurate division of DR Levels, a DR Recognition and classification algorithm based on ResNet and transfer learning is proposed. Firstly, shallow feature extraction module of ResNet18 is used to get retinal image feature, and then the fully connected classification structure model of DR Is designed. Then the transfer learning method is combined to train the network weights to improve the generalization ability of the model, ResNet-18 is selected as the backbone network model for feature extracting. Results show that the accuracy of the training set reaches to provide useful guidance for DR Automatic diagnosis, and effectively alleviates the problem of low accuracy of DR Classification

Under the background of digital information age, faced with the increasing data scale and complexity, the application limitations of traditional centralized retrieval services are becoming more and more obvious, and it is urgent to improve the data structure expansion, incremental update control and retrieval operation efficiency. In this paper, the efficient retrieval algorithm and technology of massive data information are taken as the research object, and a set of construction scheme of big data storage and retrieval system is proposed for unstructured data, which promotes the organic combination of distributed technology and full-text retrieval technology and realizes the optimization of fast retrieval processing mode of large-scale data. The system is based on Hadoop framework, with Hbase as the data storage module, and combined with ElasticSearch engine, IKAnalyzer word breaker and Redis cache to complete real-time and efficient data retrieval. Finally, based on Java web technology, a network application program convenient for users to operate online is formed. Practice has proved that the system has solved many problems in the process of collecting, storing and retrieving massive unstructured text data. At the same time, it improves the sharing transmission efficiency and concurrent access control ability of data information, and opens up a brand-new big data retrieval service model.

With the development and maturity of UAV technology, its application fields are more and more extensive. In the actual flight process of UAV, itis necessary to calculate the safe path efficiently and reliably according to the external environment information. However, the existing UAV path planning algorithm has some limitations such as single applicable type, slow convergence speed, and easy to fall into the local optimum. The application of snake optimization algorithm in UAV path planning can solve the above problems. It can carry out fast and efficient path planning in complex environments, and the application of this algorithm provides a new direction for the problems related to the path planning of unmanned aerial vehicles.

The current conventional collaborative resource scheduling algorithms in heterogeneous cloud environments mainly process the allocation through the quantified results of data characteristics of heterogeneous cloud resources, which leads to low integrated scheduling efficiency due to the differences in the attributes of resources. In this regard, a collaborative and comprehensive resource scheduling algorithm in heterogeneous cloud environment is proposed. Firstly, the heterogeneous cloud resource information data is sampled and processed, and the resource quality is graded. The scheduling task model is constructed by constructing the mapping function of scheduling task assignment sub-nodes, and the hierarchical scheduling strategy is proposed by combining with ant colony algorithm. In the experiments, the designed collaborative integrated scheduling algorithm is tested for the scheduling efficiency. The final results can prove that the algorithm has a lower average delay and a more desirable integrated scheduling efficiency when the proposed method is used for scheduling heterogeneous cloud resources.

Aiming at the current problem of unsatisfactory vehicle detection in complex scenes, an improved vehicle target detection network model is proposed. First, Res2Net residual network is fused in SCP, and the CSP_R structure is proposed, so that the model can extract deeper feature information and strengthen the ability to characterize small-scale targets; the attention mechanism is introduced, and the C3_CBAM module is designed to strengthen the attention to the detection targets while avoiding the increase of the model's computational volume; the loss function of the MPDIoU regression optimization is introduced, and the loss function is optimized by combining the prediction frame with the real frame length, width and area loss, and quantitative indicators to improve the convergence speed and robustness of the model. Finally, the model is validated on the SODA10M dataset, and the experimental results show that the model detection speed reaches 32 frames per second. The average detection accuracy reaches 83.7%, which is an improvement of 7.8 percentage points compared with YOLOV5s.

3D reconstruction is 3 d reconstruction in aerial reconstruction, industrial measurement, medical image reconstruction, cultural relics preservation and restoration, virtual reality and other fields. However, the traditional 3 D reconstruction algorithm will have a poor reconstruction effect due to the transient foreground influence and the limitation of feature point identification in the scene. To solve the above problems, this paper uses LoFT R-SIFT algorithm to extract the feature points in weak texture area, increase the number of feature points matching in weak texture area, then introduces ExtremeC3Net algorithm to eliminate the feature points on the dynamic portrait in the scene; Finally, DPT improves the MVS algorithm to make deep compensation. The experimental results prove that the feature point matching accuracy of the algorithm is improved by 55%, which can better capture the details of the scene

The traditional system capability assessment methods are no longer able to address the challenges in system evaluation. Further development and innovation in algorithms are needed. This paper addresses the new challenges encountered in the system design process and proposes a system capability assessment algorithm centered around the evaluation of capability measurement values, capability advantages and disadvantages, as well as capability improvement and decline. These provide a scientific basis for determining the focus, direction, scale proportion, and improvement optimization of the system.

The detection data of energy consumption and carbon emissions may be affected by equipment failure, sensor error, incomplete data collection and other factors, resulting in low detection accuracy. Based on this, an online detection method of energy consumption and carbon emissions measurement based on hybrid particle swarm optimization algorithm is proposed. Analyze the measurement principles of energy consumption and carbon emissions. On this basis, collect the measurement data of energy consumption and carbon emissions in real time, use semi-supervised learning to extract the measurement operation data, calculate the Angle between the newly generated optimization solution and the reference direction vector, and use it as the attribute space of particle update, and assign all optimization target values, and use the hybrid particle swarm optimization algorithm. The on-line measurement process based on hybrid particle swarm optimization algorithm is completed. The experimental results show that the proposed method has advantages in all aspects of performance index, AUC value is higher than 0.9, detection time is lower than 8s.

The development of information technology has made the field of deep learning face recognition develop rapidly. The traditional face detection and recognition algorithm can perform well under constrained conditions, but under unconstrained conditions, its effect will be greatly discounted when low quality images and partial occlusion of faces are encountered. Based on MTCNN and FaceNet, this paper adopts two strategies to solve the above two problems respectively. On the one hand, by introducing the face image quality assessment function to solve the problem of low quality pictures, before face detection, a quality assessment of the face image is done, and only the image whose quality score reaches the threshold can be input into the model. On the other hand, the Coordinate attention mechanism is introduced to deal with the problem of partial occlusion of the face, which improves the recognition ability of the model by adaptively enhancing the weight of the unocclusion area of the face. Experimental results show that compared with existing algorithms, the accuracy of the proposed algorithm is significantly improved.

To address the low efficiency of manual inspection methods used for Citrus Huanglongbing prevention and control, a system design of citrus huanglongbing in-field detection with AI edge computing device is proposed and evaluated. The system consist of Image Capture Robotic Devices, AI Edge Computing Service, Cloud Service, and Remote Control Client. A citrus Huanglongbing detection neural network model was trained with 84.1%mAP, which can be deployed on an AI edge computing device, such as Jetson Nano to detect HLB with lower delay than using a cloud-based AI approach. Therefore, robotic devices such as UAVs, surveillance cameras can be used to efficiently inspect citrus orchard, process images of citrus leaves collected from cameras in real-time. Experimental result shows that this system has great potential to apply on Citrus Huanglongbing field detection scenario to enhance the inspection efficiency of citrus orchards.

As the global pandemic gradually eases and the aviation transport industry continues to experience steady growth, highdensity flight operations are becoming the new normal. The intelligentization of flight support processes is a crucial avenue for enhancing both the safety and efficiency of flight operations. With the advancement of computer vision technology, video-based object tracking has shown significant potential in the context of flight support processes. However, in real airport environments, object tracking often encounters challenges such as occlusion, scale variations, rotation, and changes in lighting conditions, leading to a decrease in tracking accuracy and even target loss. In this paper, our focus is on overcoming tracking failures caused by occlusion, deformation, and lighting variations. We have conducted the following work, taking into consideration the unique characteristics of airport environments and the specific requirements of flight support processes: (i) We utilized features at three levels, namely, Histogram of Oriented Gradient (HOG), Color Names, and Convolutional Neural Networks (CNN), to describe the texture, color, and high-level semantics of video images, respectively. (ii) We employed a multi-feature fusion approach using a trilinear interpolation function to integrate information from various sources. (iii) We implemented improved ECO algorithms for the tracking of moving objects in the airport environment. Finally, we validated this object tracking system using real surveillance videos from the airport. Experimental results have demonstrated the effectiveness and practicality of the method under challenging conditions.

Reversible Data Hiding in Encrypted Images (RDHEI) embeds information while protecting the content of images from being leaked, allowing users to decrypt image content, extract embedded information, and losslessly recover the original content based on the key types they possess. It is a recent hot research area at the intersection of information hiding and encrypted computation, aiming to ensure both data security and the ability to hide information within images. However, inadequate utilization of image blocks in RDHEI results in a low embedding capacity of additional data. For this reason, this paper proposes a RDHEI based on adaptive difference prediction and block subdivision. At first, divide the image into equally sized blocks, and these blocks are encrypted to conceal the content of the image. For data hider, using adaptive the most significant bit(MSB) prediction to classify the available and unavailable blocks. Based on adaptive MSB prediction(AMP), adaptive difference prediction(ADP) is used to subdivide the unavailable blocks to vacate more room for data embedding. When receiver receives the embedded encrypted image, the embedded data or image can be decrypted according to its own key possession. Experimental results show that the proposed method has a significant effect on improving the embedding capacity.

The FBS_YOLO3 vehicle detection algorithm is a novel solution to the challenge of detecting vehicles in unstructured road scenarios with limited warning information. This algorithm builds upon the YOLOv3 model to deliver advanced multi-scale target detection. Firstly, FBS_YOLO3 incorporates four convolutional residual structures into the YOLOv3 backbone network to obtain deeper feature knowledge via down-sampling. Secondly, the feature fusion network is improved by implementing a PAN network structure which enhances the accuracy and robustness of viewpoint recognition through top-down and bottom-up feature fusion. Lastly, the K-means clustering fusion cross-comparison loss function is utilized to redefine the anchor frame by employing a K-means fusion cross-ratio loss function. This innovative approach solves the issue of mismatching the predetermined anchor frame size of the YOLOv3 network. Experimental results demonstrate that FBS_YOLO3 on a self-built dataset can improve mAP by 3.15% compared with the original network, while maintaining a quick detection rate of 37 fps. Moreover, FBS_YOLO3 can accurately detect vehicles, identify viewpoint information, and effectively solve the warning information insufficiency problem in unstructured road scenarios.

It is difficult to extract deep semantic features for English composition scoring methods based on artificial features, and it is difficult for English composition scoring methods based on neural networks to extract shallow features such as the number of words, resulting in the limitations of different composition scoring methods. Based on existing research results, this paper proposes an English composition scoring method that combines artificial feature extraction methods and deep learning methods. This method uses artificially designed features to extract shallow features at the word and sentence levels in the composition, draws on existing methods to extract semantic features of the composition, and performs regression calculations on the deep features and shallow features to obtain the total score of the composition. The experiment uses the Pearson evaluation index to measure the correlation between the predicted total score of the essay and the true total score under the combination method. The experiment shows that compared with the average results of 0.747 and 0.645 of baseline models such as BiLSTM and RNN, the algorithm proposed in this article is respectively improvements are 0.068 and 0.17, which proves the effectiveness of the method proposed in this paper.

The invoice reimbursement process is very cumbersome and requires manual entry of key information in the invoice, which wastes a lot of manpower and time. Therefore, it is particularly important to design an algorithm for intelligent identification of invoice information. Traditional algorithms can identify information from scanned invoice images. However, since in our country, most of the invoice information is Chinese characters, the current recognition algorithm has a certain degree of difficulty in identifying Chinese characters, and garbled characters will appear. Therefore, this article combines the CTPN text detection algorithm with the DesNets text recognition algorithm, and uses this algorithm to detect and recognize text on the information extracted from the invoice area image. Experiments show that the model outperforms the comparison model, with a recognition accuracy of up to 99.79%.

Automated vehicle driving requires a heightened awareness of the surrounding environment, and detecting targets is a crucial element in reducing the risk of traffic accidents. Target detection is essential for autonomous driving. In this paper, we improve the CenterPoint 3D target detection algorithm by introducing a self-calibrating convolutional network into the 2D backbone network of the original algorithm. This enhancement improves network extraction speed and feature extraction capability. Additionally, we improve the two-stage refinement module of the original algorithm by extracting feature points from the multi-scale feature map rather than the single-scale feature map. This approach reduces the loss of small target feature information, and we build a data enhancement module to increase the number of training samples and improve the network model’s robustness. We validate the algorithm on the KITTI dataset and analyze domestic data visualizations. Our results show that the bird’s-eye view mAP detection accuracy of the algorithm when the target is a vehicle has improved by 1.68%, and the 3D target mAP detection accuracy has improved by 1.02% compared with the original algorithm.

The rapid development of the cross-border e-commerce market has led to an increase in logistics complexity, and intelligent monitoring and early warning systems are needed to meet the challenges. The objective of this study is to design and implement a cross-border e-commerce logistics monitoring and early warning system based on improved genetic algorithms to enhance the reliability of transportation quality. The system collects data related to cross-border e-commerce logistics transportation quality, analyzes and optimizes the improved genetic algorithm in one system, and uses the improved genetic algorithm for decision-making and planning. The system has a real-time monitoring function to discover potential transportation quality problems and conduct predictive analysis to identify the min advance for timely warning. The system can provide cross-border e-commerce enterprises with more efficient logistics and transportation quality management, reduce costs and improve customer satisfaction. It helps enterprises to cope with logistics challenges, provide more reliable services, and promote the continuous development and prosperity of cross-border e-commerce.

In view of the problems such as unclear target direction, low search efficiency, and slow convergence speed of the basic ant colony algorithm in AGV two-dimensional path planning, an improved ant colony algorithm based on artificial gravity field and triangle pruning method is proposed. The algorithm first uses the attractive strength provided by the gravity field to construct heuristic information, enhancing the guidance of the target point on the planning direction and improving the directionality and search efficiency. Then, based on the concentration enhancement mechanism of the elite ant model's pheromone, an adaptive reward update mechanism for increments is proposed to improve the convergence speed. Next, an adaptive adjustment mechanism of the pheromone heuristic factor value correlated with the iteration number is discussed to balance the randomness and search efficiency of the entire planning process. Finally, the triangle pruning method is applied to global path optimization based on global path planning, effectively reducing the number of turning nodes and improving the actual motion efficiency. Comparative experiments on path planning in two-dimensional static maps using matlab validate the effectiveness of the improved algorithm in AGV global dynamic path planning.

In order to understand the multidimensional situation prediction of digital twin active power grid, a research on multidimensional situation prediction of digital twin active power grid based on LSTM algorithm is proposed. In this paper, firstly, a multi-dimensional situation prediction algorithm of power grid key indicators based on LSTM is established to realize the change prediction of key indicators attributes of digital twin active power grid. Secondly, the data of several key indicators such as load characteristics are collected, and a multi-dimensional system prediction model is established, which can control the state of active power grid; The LSTM prediction algorithm is proposed to fit the characteristics of multi-dimensional data, and the next stage of multi-dimensional data prediction is mapped to the power digital twin, so as to realize the synchronous implementation and intelligent regulation of smart energy system operation planning. Finally, a simulation test model is established, and an example shows that the multi-dimensional situation prediction method of digital twin power grid based on deep learning can better predict and distinguish the power grid situation, and provide decision support for accurate planning of energy system in the future.

The application of autonomous mobile robots is becoming more and more extensive. Path planning is one of the core problems, and the advantages and disadvantages of path planning algorithms directly affect the movement performance of robots. This paper aims to compare the path planning performance of autonomous mobile robots based on the ROS platform using multiple algorithms such as A*, Dijkstra, RRT, PRM, including path length, execution time and stability of robot posture through experiments. The results show that the PRM algorithm generally performs well in planning efficient and stable robot paths

Most floating wind turbine foundations consist of single or multiple columns, which are prone to vortex-induced motion (VIM) under the action of uniform flow. VIM is the main reason causing fatigue damage to mooring structures. Exploring the motion response of VIM of the spar-type wind turbine structure, the spar wind turbine foundation was designed as a prototype, and the model was simplified to a cylinder. In this paper, we conducted an experimental study of its model in the circulating water tank based on Matlab image processing technology. The VIM characteristics of the model were discussed by capturing its optical measurement points, and then the structural response amplitude was obtained by using the target tracking method. The study showed that the optical measurement method could effectively monitor the motion response of the structure, and the binary method could better obtain the VIM characteristics of the structure. Meanwhile, it was found that the response amplitude of the cylindrical VIM presents four stages: initial excitation branch, upper branch, lower branch, and desynchronization. It was also revealed that the maximum amplitude was reached in the upper branch, and the locking phenomenon was shown at the reduced velocity 4.96 ≤ Ur ≤ 7.11.

With the construction of city and the continuous expansion of power grid assets, transmission lines are increasingly radiating and expanding from urban areas to suburbs, mountainous areas, and even unmanned areas. Overhead transmission lines exposed in the wild are often susceptible to the impact of tree barriers. When the trees around the transmission line grow to a certain height, causing the distance between the wires and the trees to be too small, it can cause the wires to discharge from the trees, leading to accidents such as short circuits and trips. Therefore, the investigation of tree barriers is a highly concerned issue for various provincial companies. Based on the advantages of satellite remote sensing, such as wide coverage and unrestricted environmental conditions, this article proposes a three-dimensional monitoring method for tree obstacle based on satellites remote sensing images that integrates tower features. The effectiveness of the proposed method in this paper is verified by conducting experiments on a 500 kV transmission line in Chongqing and comparing it with unmanned aerial vehicle monitoring methods.

In order to understand the obstacle distance measurement method of UAV, a research on obstacle distance measurement method of UAV based on image processing technology is proposed. This paper proposes an obstacle detection algorithm based on edge extraction. The outline of the obstacle can be obtained by extracting the edge features in the image, and the size, shape and position of the obstacle can be obtained by using a rectangle to frame the outline. Then the distance of the obstacle can be extracted from the contour of the obstacle by using the depth map generated by the binocular camera. The millimeter wave radar in front is used to fuse ranging with the central area of binocular camera image to improve the update frequency of obstacle distance. Finally, the effectiveness of the obstacle avoidance control strategy is verified by the obstacle avoidance flight test, and the UAV can choose the optimal obstacle avoidance direction and successfully bypass when encountering obstacles. The comparison of obstacle avoidance results shows that the obstacle avoidance method in this paper is advanced and has certain engineering application value.

Coal gangue sorting is an important link in the process of coal mining and processing, which can effectively reduce the difficulty and cost of coal post-processing. Aiming at the problems of complicated sorting process and low sorting efficiency of coal gangue, a coal gangue sorting method based on deep learning was proposed. The method is based on the YOLO v7 deep learning algorithm, and it achieves real-time detection of coal gangue by creating a coal gangue dataset and training the detection model. By constructing a coal gangue sorting platform, the capture of target gangue has been achieved. The experimental results show that the mAP of YOLO v7 model is 96.70%, and the detection speed is 69fps, which has significant advantages compared to YOLO v5, SSD and Faster RCNN algorithms.

The rapid development of artificial intelligence facilitates the improvement of image processing algorithms. For an intelligent inspection robot, the ability to analyze the environment through image collection plays an important role. It needs to collect multiple images of the same scene from different angles of view so as to make a thorough analysis about the environment it locates and generate further decisions. Therefore, a technique called image stitching is used. Currently, the development of image stitching algorithms is getting mature – multiple algorithms have already been proposed based feature extraction techniques. However, these existing algorithms are usually unable to handle the problem of parallax existing in real world image. Therefore, in order to solve it, we proposed an unsupervised deep learning image stitching algorithm, which uses infrared images to provide auxiliary information. We utilized our own equipment to collect real world images in visible light and infrared. Finally, we implemented our own model and other popular existing image stitching algorithms and compared and contrasted their performance on our dataset. The results showed that our model has the best performance in all aspects than other algorithms on the dataset, indicating the strong advantages of deep learning methods on image stitching tasks

It is crucial to ensure the safety of personnel and prevent unauthorized intrusion in the non-stop construction area of large airports. This study proposes an image recognition method for dangerous behavior of non-stop construction personnel in large airports based on infrared imaging technology. Using infrared imaging technology to collect visual information of images of non-stop construction personnel in large airports, and analyzing images using structured similarity features; Based on supervised comparative learning, the method of extracting backbone features is adopted to achieve dynamic feature segmentation and reconstruction processing; Based on ambiguity analysis, extract the edge bounding contour features of personnel and identify dangerous intrusion behaviors of personnel. Through experimental verification, this method has high accuracy in detecting personnel's dangerous intrusion behavior.

To obtain higher imaging quality, the dark level generated during the operation of CMOS image sensors (CIS) needs to be corrected. In this paper, a dark level correction circuit is designed based on a 4 T active pixel, which includes a dark current cancellation circuit and a switched capacitor amplifier circuit. First, the dark current is collected in real time by using the dark pixels in the periphery of the face array, and the dark current noise is read out and differed from the image signals output from the columns to obtain a more accurate output signal, thus eliminating the dark level caused by the dark current. Then the switched-capacitor amplifier is used to collect and amplify the signals to facilitate the subsequent ADC processing. Based on the 110 nm process for the proposed method of specific circuit design verification, the verification results show that the dark level correction circuit designed in this paper through a real-time sampling of the dark pixels of the periphery of the array can be reduced to the exposure stage of the dark current noise to more than 85% of the original.

Graph model layout technology is an important cornerstone in graph visualization. Although the present graph model layout methods have been well studied, there are obvious problems: (1) excessively high initial state correlation; (2) excessive reliance on local optimal solutions; (3) limitation on the number of nodes. In this paper, we propose a new graph layout method on a graph grammar framework. First, the input graph model is parsed by graph grammar, with the reduction process recorded. Next, in the reverse order of reduction, the derivation operation starts from the initial graph and ends at a redrawn graph, with a new layout that meets the required specifications. Compared with other methods, regardless of the initial state, this method combines global and local layout specifications in productions and provides an intuitive yet effective way for the graph layout adjustment.

Image blur and detail information loss are caused by various factors such as imaging environment and hardware performance, therefore a multi-level image detail enhancement method based on guided filtering is proposed. Firstly, the input image is iteratively filtered by using the guided filter, to obtain background images with different smoothness; then the background image is subtracted from the original image to obtain detail images with different levels; finally, a dynamic saturation function is used to adjust the weights of detail images, which are superimposed with the original image to obtain the enhanced image. The proposed method is compared with the existing enhancement algorithms using open dataset. The experimental results show that, compared with other enhancement methods, the proposed method in this paper achieves a better enhancement effect, the enhanced image has clear edges, and the visual effect is suitable. Compared with other methods, the objective indicators of information entropy, average gradient, and spatial frequency are improved on average. 1.39%, 27.9%, and 19.3%.

In the process of text classification, due to the short length of news headlines and sparse text features, it is difficult for headline samples to provide sufficient and valuable features for classification learning, resulting in low classification accuracy. This paper combines TextCNN with the GRU model, and uses the extraction algorithm proposed in this paper for word embedding, and proposes the TC-GRU (TextCNN-GRU) model. In order to verify the effectiveness of the model in this paper, experiments were carried out on THUCNews, Sogou News Dataset and Fudan University Text Dataset. Experimental results show that the accuracy of the TC-GRU model is generally higher than that of other comparative models

Interactive video object segmentation (iVOS), which aims to efficiently produce high-quality segmentation masks of the target object in a video with user interactions. Recently, numerous works are proposed to advance the task of iVOS. However, their usages on user intent are limited. First, typical modules usually try to direct generate the segmentation without any further exploration on the input interaction, which misses valuable information. Second, recent iVOS approaches also do not consider the raw interactive information. As a result, the final segmentation results will be poisoned by the erroneous information given by the previous round’s segmentation masks. To solve the aforementioned weaknesses, in this paper, an Iterative Segmentation and Propagation based iVOS method is proposed to conduct better user intent exploration, namely ISP. ISP directly models user intent into the PGI2M module and TP module. Specifically, ISP first extracts a coarse-grained segmentation mask by analyzing the user’s input. Subsequently, this mask is used as a prior to aid the PGI2M module. Secondly, ISP presents a new interaction-driven self-attention module to recall the user’s intent in the TP module. Extensive experiments on two public datasets show the superiority of ISP over existing methods.

Conventional road boundary intelligent identification methods mainly use black-and-white optical driving method to generate road boundary detection binary image, which is easily influenced by threshold segmentation, resulting in large deviation parameters of parabola identification. Therefore, it is necessary to design a brand-new road boundary intelligent identification method based on image segmentation and edge features. That is to say, the road boundary is extracted by using the image edge features, and an intelligent road boundary identification algorithm is designed in combination with image segmentation, thus completing the intelligent road boundary identification. The experimental results show that the intelligent road boundary recognition method based on image segmentation and edge features has good recognition effect, reliability and certain application value, and has made certain contributions to improving driving safety.

Based on the color model, this paper investigates the green glass in Cantonese colored windows, aiming to establish the standard value of the green color of Cantonese colored windows, and to provide relevant data and suggestions to protect the design concept of Cantonese architectural decoration and regulate the use of color in colored windows. By collecting samples, with the help of image processing and color analysis software, the green glass was positioned and analyzed in both RGB and LAB color models. It is found that in the RGB color model, the threshold intervals of green glass are mainly concentrated in the Forest Green and Green regions; in the LAB color model, the threshold intervals of the two-color channels of green glass are mainly distributed in the range of medium and low saturation. This study provides digitalized standard values and reference data for the green of the Cantonese colored windows, which helps to maintain the design style of traditional architecture and promote the development and application of colored windows.

This paper proposes a fine-grained image classification architecture using multi-task learning. The structure of the fine-grained classification network uses ResNest as the feature extraction layer of the multi-task hard parameter sharing mode with the fine-grained category label regression branch based on multi-hot naming conventions and classification branch based on cross-entropy loss with one-hot encoding. The coupling between the two branches enables multi-task classification through hyperparameter weighting. Subsequently, comparison and ablation experiments were performed on the public datasets of Stanford Cars, CUB-200-2011 and FGVC-Aircraft. The experimental result shows multi-label regression, multi-task learning and label smoothing can effectively improve the generalization of the model and increase the inter-class distance of the previous layer at the network output terminal, and reduces the intra-class distance.

To achieve automatic fruit object recognition in complex backgrounds, this paper proposes a fruit object detection algorithm based on YOLO-GF. Addressing challenges such as complex backgrounds, significant variations in target shapes, and instances of occlusion in fruit images, we utilize the Global Attention Mechanism (GAM) to enhance the feature extraction capability for fruit targets, thereby improving fruit recognition accuracy. Additionally, the Focal-EIOU loss function is used instead of the CIOU loss function to expedite model convergence. Experimental results demonstrate a significant improvement in recognition accuracy under the same hardware conditions. On the same test dataset, the improved model achieves an mAP₅₀ of 92.1% and mAP_50:95 of 76.5%, representing increases of 5.8% and 11.9% compared to the original model, respectively.

In view of the fault and leak detection problems caused by complex scenes of offshore area in remote sensing image ship detection, a lightweight ship classification detection method is proposed based on improved YOLOv7-tiny. On the one hand, this method stacks a lightweight feature extraction module and applies it to the backbone feature extraction network, which significantly reduces the parameter and computational complexity and does not weaken the network's ability of feature extraction. On the other hand, this method introduces spatial information into the feature pyramid, raising the discrimination of features at different scales, to improve the classification and detection ability of the network. This method has been tested on the remote sensing image ship data set. The experimental results show that the average accuracy of ship classification detection based on the improved network is increased by 2.9%. Meanwhile, the parameter quantity and computational complexity are better than YOLOv7-tiny, with a 15% reduction in parameter quantity and a 24% reduction in computational complexity.

With the wide application of 3D building cluster models in urban planning, visualization and other fields, how to improve the rendering efficiency and reduce the computational cost of building cluster models has become an important issue. To address this problem, this paper proposes a visual perception evaluation model used to assess the weights of buildings based on multi-factor considerations to determine the order of building simplification, and weights the vertex importance for the classical QEM algorithm to redefine the collapsing cost of the edges, which achieves the purpose of reducing the complexity of the model while maintaining the visual quality. Experimental results show that the algorithm can significantly reduce the model rendering time and computational cost while maintaining the visual quality.

Based on 3D images, this study aims to explore automatic segmentation and enhancement methods for airfield runway surface cracks. Firstly, a typical 2D Gaussian filter is used to remove noise from the road surface data. Then, Steerable Matched Filter (SMFB) is introduced to extract crack features. By constructing a set of 52 SMFB filters with different parameters, we are able to accurately capture cracks with different directions and sizes. After that, Tensor Voting (TV) technique is introduced to further enhance the continuity of the cracks. With this method, we are able to detect and segment the cracks in the airfield runway surface for a more accurate and comprehensive analysis. The experimental results show that the proposed method performs well in crack detection and segmentation, providing strong support for airport pavement maintenance and management.

Underwater imaging is plagued by light absorption and scattering, resulting in distorted, blurry, and low-contrast. This paper introduces an innovative underwater image restoration algorithm that combines natural lighting-based airlight estimation with the refined dark channel prior. The algorithm directly estimates airlight, considering various underwater conditions such as depth, water quality, and camera-object distance, using the Jaffe-McGlamery underwater image formation model tailored for real-world underwater scenarios. A transmission map formula rooted in the refined dark channel prior is then derived. Finally, the algorithm employs the estimated airlight and transmission map to restore the image. Experimental results validate the algorithm's effectiveness in removing airlight artifacts, enhancing image contrast, and providing a clearer and more natural visual output. This approach promises to advance the quality of underwater imaging and its applicability across various domains.

This paper introduces a novel approach to accelerate Magnetic Resonance Imaging (MRI) using 1-dimensional undersampling and compressed sensing. By strategically applying under-sampling to rows through a Gaussian distribution, the proposed method aims to reduce the number of samples required for image reconstruction while maintaining image quality. The reconstruction process involves denoising with a Projection Over Convex Sets (POCS) algorithm, optimizing the threshold parameter lambda (λ) for effective denoising and convergence. Simulation results showcase the method’s effectiveness. Reconstructed images at varying under-sampling rates illustrate the gradual reduction of artifacts with increased mid-frequency sampling. The study also explores different lambda settings during reconstruction, highlighting the balance between denoising and convergence. While this approach shows promise for accelerating MRI and other imaging applications, challenges include evaluating alternative "mask" matrices and exploring under-sampling patterns beyond Gaussian distribution. The paper concludes by emphasizing compressed sensing’s potential to enhance applications constrained by scan time, fostering optimism for broader adoption.

In response to the thin nature of hot rolled steel plates and strips, the vast majority of which are surface defects that can easily lead to production accidents, and limited by the challenges of insufficient datasets and a large amount of unlabeled data, this paper proposes a comparative learning method to solve the above problems. In terms of methods, a dual data augmentation strategy is adopted. Firstly, the original image is data enhanced through manual processing, and CycleGAN is introduced for style transfer to enrich the dataset. Then, ResNet152 network is used for feature extraction, and several comparative learning methods are applied to observe the accuracy of hot rolled strip defect detection. In the end, the improved comparative learning method in this article successfully improved the accuracy of surface defect classification for hot rolled strip steel. Through this research, we are committed to providing more reliable quality control methods for industrial production and reducing the risk of production accidents.

Starting from the problem that rice extraction from remote sensing images still faces effective feature construction and extraction model, the feature optimization and combined deep learning model are considered. Taking Sentinel-2A image as data source, a multi-dimensional feature data set including spectral features, red edge features, vegetation index, water index and texture features is constructed. The ReliefF-RFE algorithm is used to optimize the features of the data set for rice extraction, and the combined UPerNet-Swin Transformer model is used to extract the rice from the study area based on the optimized features. Comparison with other feature combination schemes and deep learning models demonstrates that: (1) using the optimized features based on the ReliefF-RFE algorithm has the best segmentation effect for rice extraction, which its accuracy, recall rate, F1 score and IoU reach 92.77%, 92.28%, 92.52% and 86.09%, respectively, and (2) compared with PSPNet, Unet, DeepLabv3+ and the original UPerNet models, the combined UPerNet-Swin Transformer model has fewer misclassifications and omissions under the same optimal feature combination schemes, which the F1 score and IoU are increased by 11.12% and 17.46%, respectively

Pneumonia is a life-threatening respiratory infection that affects millions of individuals worldwide. Early and accurate diagnosis of pneumonia is crucial for effective treatment and patient care. In recent years, deep learning techniques have shown remarkable promise in automating the diagnosis of pneumonia from X-ray images. However, the inherent variability in X-ray images and the complexity of pneumonia patterns pose significant challenges to achieving high classification accuracy. In this paper, we propose a novel approach for pneumonia X-ray image classification based on multiple model ensemble. Our method leverages the strengths of diverse deep learning architectures and achieves superior classification performance compared to single models. We conducted extensive experiments on both public and private datasets, and the proposed method achieved accuracy improvements of 7.53 and 3.36, respectively. The experimental results indicate that the proposed method has high usability.

Remote sensing image ship target detection and course discrimination is one of the important supports for building a maritime power. Since ship target in remote sensing images are generally in strips, the IOU score is very sensitive to the angle of bounding box. Moreover, the angle of the ship is a periodic function, this discontinuity will cause performance degeneration. Meanwhile, methods generally use oriented bounding boxes as anchors to handle rotated ship target and thus introduce excessive hyper-parameters such as box size, aspect ratios. Aiming at the problem of complex calculation of anchor frame traversal mechanism and discontinuity of angle regression caused by increasing angle attribute in ship target detection of remote sensing image, a ship target heading detection method based on ship head point is proposed. The discontinuous angle regression problem is transformed into a continuous key point estimation problem, and the ship target detection and heading recognition are unified. Second, CA attention mechanism is added to the feature extraction network to enhance the attention to the ship target and predict the center point of the ship target. The offset and target width at the center point are regressed. Then, return the heading point and offset to obtain the accurate heading point position. Next, the rotation angle of the ship is determined according to the coordinates of the center point and the ship head point. Combined with the predicted width and height of the ship, the rotation frame detection of the ship target is completed. Finally, the center point and the bow point are connected to determine the course of the ship target. The effectiveness of the proposed method is verified on the RFUE and open source HRSC2016 datasets, respectively, and it also has good robustness in complex environments.

Conventional encryption and return method of UAV converter station inspection image mainly uses Fibonacci scrambling technology to generate two-dimensional encryption mapping, which is easily affected by pixel iteration, resulting in a high difference in peak signal-to-noise ratio. Therefore, a new encryption and return method of UAV converter station inspection image is needed, and an encryption and return algorithm of UAV converter station inspection image is designed based on chaotic sequence. The experimental results show that the difference between the peak signal-to-noise ratio (PSNR) before and after the encrypted return of the inspection encrypted image of the UAV converter station is small, which proves that the encrypted return of the inspection encrypted image is effective, reliable and has certain application value, and has made certain contributions to improving the inspection effect of the UAV converter station.

Embroidery is an important intangible cultural heritage in China. The development of digital technology has changed the way of transmission and inheritance of traditional culture. At present, the research on digital simulation of embroidery is still relatively small, and there are some problems such as weak generalization ability and weak three-dimensional sense. According to the characteristics of embroidery art works, this paper proposes an embroidery style generation method combining attention mechanism and cycle-consistent adversarial networks. The attention mechanism module is used to guide the generator and discriminator to control the target area migration of embroidery style images, so as to digitally simulate the embroidery art style. The results show that the proposed method has stronger generalization ability than the traditional embroidery digital simulation method, and has greater optimization in embroidery reality compared with the existing deep learning model.

Airport Collaborative Decision Making (A-CDM) is a system designed to improve the overall operational efficiency of airports. Flight support node information serves as a critical data source for the A-CDM system. Traditional manual methods of collecting node information pose several challenges, including issues related to timeliness and accuracy, making it difficult to meet the system's requirements. Some ground support operation data cannot be automatically obtained. To address this issue, this paper introduces the design concept of A-CDM time node collection using a video analysis service platform. It outlines the key technologies, components, and the implementation process of the algorithm repository, as well as discusses its future prospects. This solution aims to automatically collect data, such as milestone event timestamps during flight airport turnarounds, which are difficult to obtain manually. It brings economic and social benefits to airports and lays the foundation for smart airport development.

In view of the difficulties of volume measurement in the evaluation of refrigerator energy efficiency grade, based on the analysis of the principle of each measurement method, combined with the characteristics of refrigerator volume measurement, the technology combining structured light and binocular vision measurement was used. The structured light adopts speckle structured light. After the scanner is calibrated, the data is collected. After the data is denoised, smoothed, and edges are extracted, the volume is calculated. The results show that the structured light 3D scanning measurement accuracy meets the measurement requirements, but there are also problems such as insufficient depth of field measurement of the scanner at the turning point of the refrigerator, and the need to paste a large number of landmarks in the area without obvious features to achieve the stitching of the measured images.

The three-dimensional laser scanning technology is a non-contact measurement of objects by using a three-dimensional scanner. It is a new surveying and mapping technology, also known as real-life replication technology. This technology can completely reconstruct the surface of the scanned object through the 360-degree rotation of the laser emitter. The accuracy of point cloud data is very high. Each three-dimensional data in the laser point cloud is the real three-dimensional data X, Y, and Z coordinates of the building target, which makes the post-processing data true and reliable, and its accuracy reaches 3mm ~ 6mm. The working principle of this technology is through reverse three-dimensional data acquisition and model reconstruction, on-site photography, texture mapping for the later period, can truly achieve 1 : 1 physical restoration, and then quickly reconstruct the three-dimensional model of the building, so as to obtain the building point, line, surface, body and other mapping data. Three-dimensional laser scanning technology has played a huge advantage in the reconstruction and protection of wooden structure drum tower.

The unidirectional LSTM based music generation model does not take into account the influence of future information when generating music. It solely focuses on learning the dependencies of the current moment on past information, resulting in music with poor stability and subpar quality. To address this issue, we have developed a music generation model based on bidirectional LSTM. During the training phase, this model effectively captures musical information from both past and future time steps, resulting in a probability distribution of musical elements that closely approximates real-world music. This, in turn, leads to enhanced structural stability and improved music quality in the generated compositions. Finally, we conducted validation experiments on our proposed approach, and the results unequivocally demonstrate its effectiveness.

In the rapidly advancing information technology era, information overload poses a significant challenge. Recommender systems offer a partial solution, yet traditional methods grapple with issues like sparse data and accuracy. For this reason, this paper introduces a novel approach—a high-order graph convolutional collaborative filtering model. This model employs a subgraph generation module to enhance the importance of neighbor nodes during high-order graph convolutions. Our approach yields enhanced embeddings by embedding user-item interaction information using graph techniques, stacking multi-layer graph convolutional networks to capture complex interactions, and leveraging both initial and convoluted embeddings. This paper introduces a constraint loss function to address over-smoothing in graph-based recommendations. Our method's effectiveness is confirmed through extensive experiments on three real-world datasets

In underwater imagery, issues such as non-uniform illumination, blurriness, and low contrast are prevalent, significantly impacting the quality of captured images. In recent years, numerous researchers have delved into underwater image processing. Due to the intricacies of underwater environments, low-light images have different requirements compared to well-illuminated ones. However, existing algorithms often struggle to address the non-uniform illumination issues stemming from various lighting conditions in underwater settings. They also lack the capability to adaptively enhance underwater images with varying brightness. To tackle these challenges, we propose an adaptive illumination enhancement method for underwater images. This algorithm offers the capability to adaptively enhance underwater images suffering from detail blurriness based on their original brightness. Furthermore, it dynamically adjusts the parameters of the gamma function using the image's illumination component to augment color contrast. Experimental results demonstrate that our approach outperforms other algorithms, as evidenced by superior scores in UIQM metric. It effectively addresses edge blurriness and non-uniform illumination issues prevalent in underwater images captured under varying lighting conditions.

In order to solve the disassembly plan of the target parts in the product with high efficiency, a disassembly hybrid graph model of the target parts is proposed and established based on the disassembly connection relationship and disassembly priority constraint relationship between the parts in the product. The disassembly sequence planning problem of the target parts is transformed into a search and optimization problem for the path with the optimal value in the graph model. At the same time, the sorting algorithm is used to solve the mixed graph model of the target part disassembly, finally, an example is given the feasibility of this method has been verified.

Synthetic Aperture Radar (SAR) is capable of producing high-resolution complex-valued pictures, which have extensive applications in both civil and military domains. Among these applications, SAR electronic countermeasures currently represent a prominent area of research interest. Presently, within the radar electronic countermeasures, there exists a diminishing disparity among the features of real and false targets, rendering the detection of jamming increasingly challenging. This paper examines the phase of SAR images and presents a method for identifying SAR jamming regions based on phase features. The initial step involves organizing the cluttered phase information into neighborhood phase differences. Subsequently, this information is coupled with the amplitude to obtain the weighted phase difference. This metric effectively captures the extent of phase distortion resulting from jamming. The findings from the simulation experiment demonstrate that the proposed feature and method are capable of accurately identifying and filtering out the jamming region in SAR pictures. Furthermore, it demonstrates the prospection of phase within the SAR image interpretation and electronic countermeasures.

Functional tissue region segmentation is the segmentation and example description of tissue epithelium, glandular cavity, fiber and other tissues in the image, which helps to accelerate the understanding of the relationship between cells and tissues in the world. By better understanding the relationship between cells, researchers will have a deeper understanding of cell functions that affect human health. Based on convolutional neural networks, we combine the structural advantages of UNet and EficientNet to create an organ tissue segmentation model. The model fuses the UNet structure with the EficientNet structure, and extracts features with the help of the pre-trained EficientNet optimal structure to improve the ability of feature learning. At the same time, the fusion of multi-scale features in the network is realized through the jump connection, and the segmentation accuracy of the model is improved. We compare our model with other models using the metrics of dice similarity efficiency. our Unet2.5D (ConvNext+ Se_resnet101) owns the highest DSC 0.702 among these models, which is 0.052, 0.024, 0.052 higher than Unet(ResNet50), Unet(Se_Resnet101), Unet(ResNet101) respectively.

The development of the Metaverse nowadays has sparked widespread emotions among researchers, and correspondingly, many technologies have been derived to improve the human's sense of reality in the Metaverse. Especially, Extended Reality (XR), as an indispensable and important technology and research direction in the study of the metaverse, aims to bring seamless transformation between the virtual world and the real-world immersion to the experiential world. However, the technology we currently lack is the ability to simultaneously separate, classify, and locate dynamic human sound information to enhance human sound perception in complex noise environments. This article proposes a framework that utilizes FCNN for separation, algebraic models for positioning to obtain estimated distances, and SVM for classification. The dataset is built to simulates distance-related changes with accurate ground truth labels. The results show that our method can effectively separate, separate, and locate mixed sound data, providing users with comprehensive information about the content, gender, and distance of the speaking object in complex sound environments, enhancing their immersive experience and perception ability. Our innovation lies in the combination of three audio processing technologies and the framework proposed may well inspire future work on related topics.

(1) Background: To improve the quality control ability of cigarette characteristic cigarette filter rods, build the filter rod structure database of cigarette products, and accelerate the Digital transformation of filter rod platform technology research and development, this study proposed a research method for digital remodeling of cigarette structure.; (2) Methods: Firstly, image acquisition of the cigarette filter rod end face is carried out by combining a color area-array camera, telecentric lens, and coaxial light source to form an area-array camera testing environment; By combining a line scanning camera with a coaxial light source and a point light source, the surface of the cigarette filter is photographed and the point light source transmittance is detected; Simultaneously combining 3D laser camera scanners to establish the contour of the target; (3) Results: By using the above methods, the detection images of four types of filter rods were processed to obtain HSV image curves, grayscale images, and color histograms. These images were used for 3D model reconstruction and 15 3D feature maps were obtained; (4) Conclusions: The reconstructed 15 3D images can accurately distinguish four different filter rods, and this method can provide a reference for real-time detection during cigarette filter rod processing.

To enhance the performance of intelligent watt hour meters, a visual recognition and comparison system based on convolutional neural networks is proposed for intelligent watt hour meter chips. Firstly, the overall framework of the chip visual recognition comparison system is designed. Secondly, the hardware part of the system comprises the image acquisition module and image data transmission module of intelligent watt hour meter chips. In the software part, the classification function is selected based on the structural characteristics and operational principle of convolutional neural networks, and iterative training is used to complete the identification and comparison of smart meter chips. The experimental results demonstrate that this proposed system can significantly improve the accuracy of visual recognition and comparison, while also reducing the time consumption when compared to traditional recognition and comparison systems.

With the continuous improvement of the application of artificial intelligence theory and intelligent hardware processing capabilities, the application of machine vision and unmanned technology is becoming more and more popular. As the city scales rapidly, fire safety issues are becoming more and more important. Taking into account the safety of firefighters and the efficient handling of accidents, an intelligent unmanned fire car was designed to simulate fire safety problems in toxic and harmful environments. The functions of this smart car include: automatic driving, image recognition, monitoring of remote and environment, etc., which can help firefighters to extinguish fires in a harmful environment and minimize property damage.

The marine gas turbine propulsion system generally works in a healthy state, and the samples collected by the monitoring system are characterized by more normal samples and fewer fault samples. Aiming at the problem of lack of fault samples faced by data-driven fault diagnosis methods, a cross-working condition fault diagnosis model is proposed by using transfer learning to reduce the dependence of data-driven methods on fault samples. The proposed method was experimentally validated by using a real-ship-validated dataset. Compared with traditional methods, the proposed method can achieve cross-working condition fault diagnosis with fewer fault samples.

This paper focuses on the feasibility study of vision-based deformation monitoring. Sparse optical flow, a computer vision technique, is often used for deformation tracking and extraction. The objective of this study is to explore the application of optical flow in concrete column deformation monitoring. The performance of the technique is compared with the traditional LVDT (Linear Variable Differential Transformer) measurement method. The experimental results show that the optical flow method can effectively extract the deformation information of concrete columns with a smaller error compared with the LVDT method, whose relative standard deviation is 1.03 mm and relative error is 1.88%, confirming the feasibility and effectiveness of the vision-based approach. This study provides a visual solution for deformation monitoring of concrete columns, and provides reference and guidance for deformation analysis and structural monitoring in related fields.

Concrete will produce cracks under the long-term action of loads and affect the building safety, due to the unsatisfactory accuracy and efficiency of manual detection of concrete cracks, a concrete crack recognition and detection method based on improved CNN and DoG operator is proposed. Firstly, the recognition ability is trained by CNN to locate the valuable images from the image dataset, and then the locating crack images are greyscaled, denoised using bilateral filtering method, considering that the filtering will make the image edges blurred, the DoG operator is used to detect the completeness of the edges, and then the image is binary transformed by selecting thresholds through the one-dimensional Otus segmentation method, and the binary map is opened by the open operation, to fill in the broken parts within the cracks and protect the crack edges, and finally the length, width, and rotation angle of the crack are calculated by mapping the complete straight lines present in the crack through Hough space. The experimental results show that the method can accurately identify and detect crack features of different shapes with superior detection accuracy.

In response to the problem of low accuracy in detecting moving targets in minefield images due to indistinct target features, complex background information, and frequent occlusions, this paper proposes a deep learning-based method for minefield moving target detection. Firstly, a fully dynamic convolutional structure is incorporated into the convolutional block of the backbone feature extraction network to reduce redundant information and enhance feature extraction capability. Secondly, the Swin Transformer network structure is introduced during the feature fusion process to enhance the perception of local geometric information. Finally, a coordinate attention mechanism is added to update the fused feature maps, thus enhancing the network's ability to detect occluded targets and targets in low-light conditions. The proposed algorithm is evaluated on a self-built minefield dataset and the Pascal VOC dataset through ablation experiments, and the results show that it significantly improves the average accuracy of target detection in minefield images.

Aiming at multi-target detection in complex human-robot collaborative assembly scenes, an improved YOLOv7 algorithm is proposed. Specifically, the Wise-Intersection over Union(Wise-IoU) loss function and the BiFormer attention module are introduced to improve the recognition performance of small assembly parts. Taking a worm-gear decelerator as an example, a dataset for assembly parts recognition is made. By training the improved network in the self-made dataset, the mAP@.5 value is increased by 3.25 % and the average total loss is reduced by 0.02365. The experiment results show that the improved YOLOv7 algorithm can achieve multi-assembly parts detection in collaborative assembly.

In order to improve the accuracy and real-time performance of cotton top bud recognition, an improved YOLOv5s target real-time detection model is proposed. First, the SE module and the CBAM module in the attention mechanism are added to optimize the weight ratio of channel attention and spatial attention to improve accuracy; then the BiFPN structure of bidirectional weighted features is introduced to strengthen the fusion between high-level features and low-level features; finally, a new bounding box regression loss function EIoU is used for ablation experiments, and more position information of cotton buds can be obtained by reducing the bounding box loss. The experimental results show that, by applying the improved algorithm in the identification of cotton top buds, compared with the original YOLOv5s model, the accuracy of the C3SE-4l+BiFPN+EIoU model has increased by 7.9%, the recall rate has increased by 2.8%, and the average precision an increase of 5.7%. These improvements and optimizations provide a new idea and method, which can provide a more efficient solution for the identification of cotton top buds.

3D reconstruction technology utilizes 3D data to create models of physical objects. Cameras, laser scanners, and other sensors can be used to gather 3D data of objects, which can be processed using computer graphics technology for creating 3D models through 3D reconstruction technology. In engineering, high-precision 3D reconstruction models can substitute physical pipes for automatic measuring of pipe diameters. This paper proposes a target segmentation-based optimization method for single-frame reconstruction, which enables precise diameter measurement of pipes. Experimental results show that single-frame reconstruction, based on target segmentation technology, produces excellent results in the current application scenario. The proposed method is better adapted to complex construction conditions than the complex reconstruction methods. Complex backgrounds include excessive and uneven distributed light and interfering objects. Using target segmentation technology based on image processing, the MIVOS user-interactive video can produce and distribute the target object mask based on the user's interaction with the video frame. Complex background removal can improve the quality of reconstructed sample images. MIVOS is used to segment the pipe area in the image and remove most of the background noise. Consequently, the process lessens the interference of background noise in the reconstruction results. The proposed method exhibits significant progress in measuring both the inner and outer diameters of pipes when compared to both multi-frame and single-frame reconstruction methods. Their measurements have an average error of no more than 1 mm. The proposed method provides technical guidance for measuring the inner and outer diameters of pipes under complex conditions.

In order to address the drawbacks of using offline and manual methods to measure the length of reconstituted cut tobacco using thick pulp method, an online automatic detection device is designed in the paper. A cut tobacco sampling and spreading out process is designed. Cut tobacco sampling is completed by using a robotic arm. Cut tobacco feeding is completed by simulating the manual shaking movement. A two-stage conveyor belt keeps spreading out tobacco leaves. A tobacco leaves image acquisition device have been designed to collect diluted cut tobacco. A software system has been developed, specifically for the case of a small amount of cross cut tobacco. A stitching algorithm based on sub pixel skeleton contour slope has been proposed to separate a single cut tobacco and facilitate tobacco length detection. The test results indicate that the system has the advantages of automation, online detection, and high measurement accuracy, which can meet the requirements of the production process of reconstituted tobacco leaves.

With the continuous improvement of intelligent management level in red bayberry orchards, the demand for automatic picking and automatic sorting is becoming increasingly apparent. The prerequisite for achieving these automated processes is to quickly identify the maturity of red bayberries by object detection. In this study, we classified red bayberry into 8 levels of maturity and achieved an object detection precision of 88.9%. We used a fast object detection model, combined with small object optimization methods and small feature extraction layers to get higher precision.

In this paper, we research the multi-objective vehicle routing problem with time windows under uncertainty. For solving it efficiently, the robust multi-objective particle swarm optimization incorporates the simulated annealing algorithm is proposed. The new algorithm aims to improve the local search abilities of particles. Experimental results show that the proposed algorithm outperforms the traditional the robust multi-objective particle swarm optimization algorithm on the selected problem sets as the uncertain interference intensity increases.

For complex stroke rehabilitation scenarios, visual algorithms, such as motion recognition or video understanding, find it challenging to focus on patient areas with slow motion amplitude and pay more attention to targets with drastic changes in light flow. Therefore, it can provide critical perspectives and adequate information for the above visual tasks using a semantic segmentation algorithm to capture the patient's area from the captured image. Currently, the weakly supervised segmentation algorithm based on bounding boxes tends to utilize existing image classification methods. They can perform secondary processing on the internal images of boxes to obtain larger areas of pseudo-label information. In order to avoid the redundancy caused by algorithm concatenation, this paper proposes an end-to-end weakly supervised segmentation algorithm. In this method, a U-shaped residual module with variable depth is designed to capture the deep semantic features of images, and its output is integrated into the target matrix of the NCut problem in the form of blocks. Then, the region of the target is indicated by solving the sub-minimum eigenvector of the generalized eigensystem, and the segmentation is realized. We conducted experiments on the PASCAL VOC 2012 dataset, and the proposed method achieved 67.7% mIoU. On the private dataset, we compared the proposed method with similar algorithms, which can segment the target area more intensively

Lung cancer is the disease with the highest incidence rate and mortality of cancer in China, which seriously threatens human life safety. Pulmonary nodules are the main factor leading to lung cancer, and their precise identification plays a crucial role in clinical diagnosis. This paper proposes a lung nodule detection model that combines global image information to address issues. The model is based on improved YOLOV5 network. Finally, comparative experiments have verified the accuracy and effectiveness of this model.

Although the Neural Radiance Fields (NeRF) has been shown to achieve high-quality novel view synthesis, existing models still perform poorly in some scenarios, particularly unbounded scenes. These models either require excessively long training times or produce suboptimal synthesis results. Consequently, we propose SD-NeRF, which consists of a compact neural radiance field model and self-supervised depth regularization. Experimental results demonstrate that SDNeRF can shorten training time by over 20 times compared to Mip-NeRF360 without compromising reconstruction accuracy.

Face recognition technology is one of the popular research directions in computer vision in recent years, which is widely used in our daily life. Therefore, this paper takes the Yolov5 algorithm as the core, introduces the COCO dataset, and at the same time introduces the Yolov5 system structure and analyzes the algorithm in terms of implementation and performance. Experiments are conducted on the detection of two targets with different genders, and by changing three different hyperparameters (number of training rounds, batch size and image size), we observe the influence of the change of different hyperparameters on the experimental effect and derive the suitable size of different hyperparameters

In order to ensure the power supply reliability of the power system, most of the power line maintenance operations use live work to complete wire breaking, wiring, replacement of fall insurance and other work. The distribution network live working robot can keep the operator away from the dangerous environment, ensure the safety of personnel, reduce the labor intensity of the operator, improve the work efficiency of live working, and has a broad application prospect. In order to work safety, more and more power technicians use the way of teleoperation robot arm to carry out live work. In this paper, a virtual reality synchronization system combined with binocular cameras is proposed. By reconstructing the live working environment in the virtual scene and combining with the remote operation mode of the mechanical arm, the system provides a new working mode for the electric power operators, so that they are no longer limited by the field of vision, and thus can complete the live working task more flexibly.

CNN network classes require multiple images per class to train. This makes facial recognition using CNN imprac- tical, as it is often hard to obtain a sufficient number of images of one person. Siamese Networks, on the other hand, uses oneshot learning, meaning that only one input image will be needed to train the network for each person. We build a facial recognition system using Siamese Network. In Siamese Networks, a single image of one person is input, and the network will learn to recognize the person by learning the embedding of the image. The embedding is used to compute a similarity score – similar images will have higher similarity scores. Another image will then be input to the same network, and the system will compare two embeddings to determine whether they contain the same person, giving a true or false output. Using the ORL and LFW dataset, we performed several experiments on multiple aspects of the Siamese Network. We experimented on the Random Erasing function for our augmented data to test the reliability of the network in facial recognition. Results show significant improvement on model accuracy for model trained on random erasing masking. This kind of facial recognition systems is versatile and can be applied to numerous use cases. For example, this kind of system can be used to provide facial recognition for persons with disability that manifests in the deformation of facial features.

Rice is susceptible to mold and mildew during storage. Metabolites such as aflatoxin produced during mildew will do great harm to consumers. To meet the need for rapid detection of normal rice adulterated with moldy rice, a rapid identification method of adulterated rice was established based on data fusion of near-infrared spectroscopy and machine vision. Using competitive adaptive reweighted sampling (CARS), genetic algorithm (GA), and least angle regression (LARS) for spectral and image feature extraction, combined with support vector classification (SVC), random forest (RF), and gradient boosting tree (GBT) nonlinear discriminant models, and use Bayesian search to optimize modeling parameters. The results show that the GBT fusion data model established by LARS optimization of spectral and image feature variables has the highest discrimination accuracy, with recognition accuracy rates of 100.00% and 98.11% for its training and testing sets, respectively. The discrimination performance is significantly improved compared to single near-infrared spectroscopy and machine vision. The results indicate that rapid identification of adulterated rice based on near-infrared spectroscopy and machine vision data fusion technology is feasible, providing theoretical support for the development of online identification equipment for adulterated rice.

The main task of pedestrian multi objects tracking technology is to continuously track multiple pedestrian objects simultaneously in video sequences and maintain their unique ID numbers. However, current pedestrian multi objects tracking models still have many problems, such as false detection, missed detection, and frequent ID number switching when pedestrians are obstructed or have overly similar appearances, ultimately leading to tracking failure. Therefore, this paper proposes a pedestrian multi objects tracking model based on TBD strategy. It mainly consists of two parts: pedestrian detector and pedestrian tracker. In terms of pedestrian detectors, this paper uses ES-YOLO pedestrian detectors. In terms of pedestrian trackers, this paper draws on the Omni-scale feature learning module in OSNet to redesign the StrongSORT pedestrian appearance feature extraction network, and ultimately obtains the StrongSORT pedestrian tracker based on omni-scale feature fusion, further enhancing its pedestrian feature extraction ability. In terms of experimental results. The experimental results of the pedestrian multi objects tracking model based on the TBD strategy in this paper on the MOT16 dataset show that the proposed pedestrian multi-objective tracking model can effectively improve the accuracy of pedestrian multi objects tracking and reduce the problem of frequent pedestrian ID number switching.

The commonly used shoreline extraction methods need to go through the generation process of point cloud digital elevation model (DEM) with large amount of calculation and easy to introduce errors. This proposed method used the change law of the coordinate value of the point at the time of obtaining the data due to the consistent acquisition method, puts forward the algorithm, and improves the accuracy and efficiency of the algorithm, so that it can quickly and accurately extract the boundary point cloud of the coastal zone, properly process the extracted boundary point cloud and transform it into the coastline point cloud in the strict sense. Compared with the measured coastline data and the coastline data extracted by isoline tracking method, the visual and quantitative data show that the extraction effect of this method is more continuous and the accuracy is higher. It can be seen from the data table that the overall standard deviation and variance of coastline extracted by this method are reduced from 0.3726 and 0.1415 of isoline tracking method to 0.1632 and 0.0266 respectively

A detection algorithm of maximum and minimum eigenvalues based on random matrix theory is proposed for the problem of abnormal detection of customer electricity consumption. Firstly, the data source matrix is constructed by time alignment and superimposed Gaussian white noise, and the sliding window method is used to obtain the window data indicating the operation status at each moment; secondly, the window data are standardized, feature extraction and other operations are performed, and the difference and the sum of the maximum and minimum eigenvalues are compared to construct the feature detection indexes and thresholds; finally, the algorithm is studied and verified by simulation. The results show that the algorithm does not depend on any model, can analyze the operation status of the system more comprehensively and adequately, and realizes the effective detection of abnormal data

Aiming at the difficulty of point feature matching in 3D reconstruction to meet the tracking requirements of weakly textured scenes, this paper proposes a visual SLAM algorithm based on grid method combining points with edge features. In the tracking thread, a method based on grid method is proposed to evaluate the feature quality of points. The textures of external environment are judged according to ORB feature description, and the information of Canny edge features of weakly textured mesh is added to improve the positioning accuracy. In the local mapping thread, the joint feature points pose and map points are iteratively optimized to improve the convergence rate of the algorithm. The simulation results show that the proposed algorithm has a good location and tracking effects in the weak texture scene.

As an image retrieval problem, person re-identification (Re-ID) relies on robust features extracted by convolution neural models. Most current methods use large backbone models for feature extraction (e.g., ResNet50). However, these large backbone models have many parameters, which cause many problems when embedded in smart camera devices. For example, the device's computing resources are limited, the real-time operation speed is limited, etc. So it is necessary to construct models with low parameters and low complexity. This paper proposes a new lightweight baseline for Re-ID, which is SCL-net and all underlying modules of the model are reconstructed. In our work, we design a new convolution unit----symmetrical combination units (SC-unit), which construct features map of richer channels by reusing feature maps from different convolution layers. In addition, we redesigned all the base modules of SCL-net and proved the effectiveness of all modules. We joint training of shallow and deep features of the model respectively to improve the accuracy of the model. Our SCL-net has about 2.3M parameters, and it can achieve 95.2%/85.9% on Rank-1 and mAP without any pretraining.

Due to the low accuracy of collecting vehicle position information, the error in the positioning stage is relatively large. Therefore, the collaborative positioning of intelligent vehicle aided navigation based on computer vision technology is proposed. Taking the computer vision equipment-smart cameras VOF/VOF-S as a specific data acquisition device, and combining with the specific running state of the vehicle, the specific parameters in the data acquisition stage are set differently, so as to realize the accurate acquisition of vehicle position information. In the positioning stage, the plane where the wheel is located is taken as the road plane, and the coordinate parameters of data information collected by several road ground points in VOF/VOF-S computer vision technology device are integrated to realize the transformation of vehicle position information in real space. In the test results, the positioning error of vehicle position under different driving conditions is always stable within 1.50m, which has high accuracy.

The operational level and service life of oil-immersed bushings are related to the moisture content in their cellulosic insulation layers. The authors proposed a novel method based on Taylor series modeling and grey relational analysis to assess the moisture content in the cellulosic insulation of oil-immersed bushings. Unlike conventional methods relying solely on frequency domain spectroscopy (FDS) data from certain frequency ranges, this method uses Taylor series to model the full spectrum and extracts a set of moisture-related feature parameters, thus avoiding dependence on specific measurement points or ranges. To establish the relationship between feature parameters and moisture levels, a database was constructed in this study. Grey relational analysis was used to propose an alternative method for assessing the moisture in bushing cellulosic insulation. Results show the average relative error between the estimated and measured moisture was less than 15.7%. Overall, the proposed method in this study provides an effective approach for assessing the moisture in the cellulosic insulation of oil-immersed bushings.

Henan Province is an important agricultural province in China, and its food production is crucial for meeting the country's food needs and ensuring food security. This article establishes a prediction model for grain yield in Henan Province based on BP neural network. Six indicators are selected as input variables, including total power of agricultural machinery, effective irrigation area, converted amount of agricultural fertilizer application, pesticide usage, sowing area of grain crops, and rural electricity consumption. Grain yield is used as output variable. The experimental results show that the error rate of the BP neural network prediction model in the training and validation stages is controlled within 3%, indicating that the model has good prediction performance and is helpful for the government to formulate agricultural planning and agricultural production management strategies.

To address the challenges of manual transmission line inspection and UAV line-following flights, this study presents a systematic UAV line tracking method based on binocular vision. The proposed method involves several key steps. First, the camera captures images of the transmission lines during flight, and these visual inputs are continually updated in realtime. Second, the images undergo pre-processing using Gaussian blur and bilateral filtering algorithms to mitigate the impact of light and noise interference on image detection. Subsequently, the 3D nodes of the line within the image are detected and extracted using the Hough transform and BM algorithm. The 3D pose of the specified tracking line is then determined using the least squares algorithm. Lastly, a visual guidance strategy is presented for the UAV to effectively track the designated line. The real-time capability and accuracy of the method are validated through experimental verification.

The electrical connection serves as a vital and abundant link in power, electronic equipment, and systems, with the electrical contact acting as its core component. In practical working conditions, fretting wear occurs during the usage of electrical contacts, leading to surface destruction and a decline in their performance. Determining the degree of wear on electrical contacts is crucial for assessing their failure in engineering applications. This study focuses on conducting fretting wear tests on copper material under different cycles for electrical contacts while utilizing machine vision algorithms to detect the morphological characteristics of wear marks. Gray threshold segmentation is applied to extract texture features from wear marks after various oxidation conditions. Pseudocolorization techniques are employed to process extracted morphologies, followed by calculating their characteristic areas. Finally, combining these results with contact resistance curves allows for judging the electrical conductivity of the electrical contact under different cycles.

The identification of launch event based on infrared image processing is an important machine learning technology for space-based early warning system. Due to the long-range detection and short duration of the powered phase of trajectory, launch event represents as a point target in the infrared image. Therefore, the main challenge is to address the issue of distinguishing the different types of infrared point target. To tackle this problem, we propose a novel method for recognizing images of rocket exhaust flame point targets at different temperatures. To validate our approach, we conducted experimental validation using simulated detecting images obtained from high-orbit infrared early warning satellites. These images are generated based on the EO/IR module in STK software. The experimental results verify the effectiveness of our proposed method in solving the challenge of infrared point target recognition.

Dung Beetle Optimizer(DBO) is an effective metaheuristic algorithm proposed in 2022. But at the same time, DBO also suffers from a local-global imbalance in the exploration process, tends to fall into local optimization and exploitability needs to be further improved, etc. Therefore, we propose an improved DBO algorithm to address these shortcomings and named it CDBO. Firstly, Tent chaotic mapping can be used for the purpose of initializing the population, improving the quality of initial solutions, promoting the enhancement of population variety, and augmenting the global search capability of the algorithm. Secondly, introducing dynamic weighting factors enables the algorithm to fully search for local areas while also taking into account global exploration. To assess the effectiveness of CDBO, a total of 12 benchmark test functions were utilized to evaluate the performance of this algorithm, wherein CDBO was compared with other widely recognized metaheuristic algorithms. The results showed that CDBO had improved search accuracy and convergence speed. Finally, CDBO was applied to airfoil optimization problem, verifying the feasibility of applying CDBO to practical engineering problems.

Combinatorial action recognition has recently attracted the attention of researchers in the field of computer vision. It focuses on the effective representation and discrimination of spatio-temporal interactions occurring between different actions and objects in video data. Existing work tends to strengthen the framework's object recognition capabilities and relationship modeling capabilities, e.g., attention mechanisms, and graph structures. We find that existing algorithms can be influenced by interaction-independent video segments in a video, misleading the algorithm to focus on additional information in the vision. For the algorithm to analyze the spatio-temporal interactions of causally related video segments in a video, a Causal Slice Recognition Network (CSRN) is proposed. This method can effectively remove the interference of video background segments by explicitly recognizing and extracting the causally related segments in the video. We validate the method on the Something-else dataset and obtain the best results.

This paper, based on the principles of brushless DC motors, constructs a mathematical model and combines genetic algorithms with traditional PID control. Genetic algorithms are used for parameter optimization to obtain the optimal solution for PID control, achieving higher control precision and stability. A simulation model of the motor and control system is developed using Simulink, and various operational conditions, including normal startup and sudden speed changes during operation, are simulated. The results show significant improvements in the motor's response speed and control precision. There is no overshoot during the startup phase, and the error during constant-speed operation is below 0.5%.