This PDF file contains the front matter associated with SPIE Proceedings Volume 12172, including the Title Page, Copyright information, and Table of Contents.

Radar intelligence has become the development trend of radar, and accurate jamming recognition plays a vital role in the intelligent perception of the radar. With the input of multi-domain signals, including time domain, pulse compression, and frequency domain signals, a multi-domain joint convolutional neural network model was developed to classify different types of radar jamming. The results in this paper showed that when the jamming-to-noise ratio of the jamming was greater than 2 dB, the recognition accuracy rate of this model was over 94%. Compared with the one-dimensional convolutional neural network model, the recognition rate of multi-domain joint model proposed here was greatly improved, which, to a certain extent, has promoted the development of radar intelligence.

Session-based recommendation is a challenging field in the research network-based behavior modeling, mainly due to the complex transfer of user interests between items and the limited information. The previous methods model the session as a sequence or a graph, which takes into account the role of time attention in session-based recommendation and achieve satisfactory performances, they still ignore the latent relationship between user interest transfer and item category. In this study, we propose a category attentive graph neural network (CAGNN) model for session-based recommendation. According to the item category, the category attention is embedded into the latent vector of the items to adaptively capture different user interests, which effectively improves the expressiveness and performance of the model. The numerical results of two real datasets show that CAGNN outperforms the state-of-the-art session-based recommendation methods.

Aiming at the problem that small unmanned surface vehicle (USV) are affected by wind waves, undercurrents and other factors affecting navigation safety, a navigation state perception system is designed. The system is equipped with high-precision six-axis attitude sensors, ship-borne smart terminals, GPRS wireless communication units and other equipment. Obtain the attitude information of the USV through the high-precision IMU six-axis attitude sensor, and provide data support for the remote operator's attitude control of the USV; calculate the heading angle through the shore-based terminal; build the ship-borne intelligent terminal through the GPRS wireless communication unit and the data transmission path of the shore-based terminal. The system satisfies the needs of USV sensing applications, and effectively realizes the detection of attitude and heading.

Protecting a driver’s privacy is one of the major concerns in vehicular ad hoc networks (VANETs). Currently, Maris et al. has proposed an efficient anonymous authentication protocol (EAAP) for VANETs. The authors claim that their scheme can avoid malicious vehicles entering into VANETs and offers conditional privacy. In this paper, we show that a malicious vehicle can enter a VANET as a legal vehicle and can successfully broadcast any messages. Furthermore, the malicious vehicle cannot be identified by a trusted authority, meaning that their scheme fails to support conditional privacy. The results of this study show that the scheme of Maris et al. is totally insecure and it can be combined with blockchain technology to design a more secure and efficient privacy protection authentication scheme in the future.

Load balance is a task that remains a challenge currently in face with the rapidly increased business volume and data flow. For the purpose of achieving the best efficiency of load balance, load distribution has recently been studied with consistent hashing algorithm in cluster load balancing system. In this work, we analyze the characteristics of load balance technology of SSL/ TLS web server cluster, design a segmentation method of consistent hashing ring based on virtual nodes and propose a distribution strategy based on dynamic weight aiming at solving the problem of service collapsing caused by the unbalanced load among microservice cluster on the basis of consistent hash algorithm. The experimental results show that the proposed load balancing strategy significantly reduces the probability of load imbalance about 25% compared with the conventional consistent hash algorithm. The proposed the dynamic allocation strategy also achieves an advantageous load balance performance for microservice distribution architecture.

In the field of military equipment knowledge, there are a large number of equipment models, weapon types, operating parameters and other time-frequency domain and airspace data information, which potentially contains a lot of valuable information. At present, in the face of these massive domain knowledge, combat related personnel cannot efficiently acquire the key knowledge, which means that they cannot provide effective guidance according to the potential key knowledge. In order to solve this problem, based on the investigation and analysis of the construction methods of the existing knowledge graph, this paper mined and extracted the knowledge of military equipment, instantiated and correlated different weapons and equipment, and constructed the knowledge graph of military equipment and question answering system. The construction of military equipment knowledge graph and question answering system can not only deeply study the key technical difficulties of domain graph, but also has strong strategic support significance for the development of related fields.

Combining fluid mechanics and traffic flow theory, this paper proposes a new macro-micro integrated simulation method, which regards vehicles driving in the city as fluid, and calculates the density and speed of traffic flow in real time through the conservation of mass and momentum of fluid mechanics. A density field, and a speed field, are formed in urban roads; When observing at a close distance, vehicles are generated in real time through the density field and speed field for microscopic simulation. When observing from a long distance, the destroyed vehicle presents a macroscopic traffic simulation. The results show that this simulation method can perform simulations very smoothly, more realistically reflecting the road status, vehicle status, and road congestion.

Distributed storage plays an increasingly important role in the context of big data. interplanetary file system (IPFS) is a distributed file system, which can form a network of all heterogeneous devices in the same way. Different from traditional HTTP protocol based on physical location, IPFS distributed network is based on content addressing and obtains files through file hash. However, this precise file search method cannot obtain files without file content hash which greatly reduces file utilization and liquidity. Therefore, this paper proposes a two-layer index scheme. After receiving the uploaded file, the node parses the file and establishes the index. The nodes are replicated using a CRDT data structure based on optimistic replication for indexing operations. IPFS pub-sub is used as the CRDT message delivery method between nodes. The first-layer index is the inverted index file corresponding to each keyword. The second-layer index is the CID of the inverted index file for each keyword. Each node maintains full index rather than through a distributed hash table stores dispersion index can ensure complete data search, at the same time greatly reduce search response time. Inverted index files are stored in IPFS network to reduce storage space and facilitate state-based replication of newly added nodes or nodes that have been offline for a long time. Finally, through the analysis of experimental data, it is proved that the scheme can greatly reduce the search response time while occupying acceptable storage space.

With the recent advance of various context aggregation approaches, remarkable progress has been achieved in semantic segmentation. However, it is still challenging to fully exploit the discriminative across-scale context information in an efficient manner. In this paper, we introduce an across-scale context attention network (ACANet) for real-time semantic segmentation. Instead of compute complex query-dependent attention map, we calculate query-independent attention map to aggregate contexts. Experimental results on Cityscape and Camvid datasets demonstrate the effectiveness of our method. In particular, our network achieves 77.4% on the Cityscape test set with a 32 FPS for 1024×2048 images on a single RTX 2080Ti GPU.

In recent years, the fraud crime rate in China is rising, which has seriously endangered the personal and property safety of citizens. Crime is the result of the comprehensive action of society, economy, politics and culture. Based on the data of urbanization and fraud rate in J Province from 2005 to 2019, through Pearson correlation analysis, it is found that urbanization has a significant impact on fraud crime rate. The artificial neural network algorithm is used to predict the crime rate of fraud, with the accuracy rate 93.7%.

The text region extraction is the key step for optical character recognition (OCR) operation. According to the layout characteristics of historical Tibetan document, this paper proposes a text region extraction method based on border detection to extract text region from documents. Firstly, the character height and stroke width are estimated and the border region position is detected. Then, the position of decorative lines surrounding the body text region is detected by heuristic search. Finally, the mask image of the document is generated according to the position relationship between the text regions and border, then different text regions are extracted. Experiments on dataset of historical Tibetan document show that this method can effectively overcome the problems of page tilt, border and decorative lines fracture, and demonstrate the effectiveness of the proposed method.

The traditional disease detection method is that a doctor judges if a user has a specific disease through the doctor's treatment based on his inference and medical detection reports provided by the medical equipment. In recent years, with the continuous development of digital healthcare, more and more users choose to conduct disease detection through a digital health system. To provide accurate disease detection service, the system needs to collect mass patients' medical information such as symptoms and signs. However, patients' medical information and users' query requests will reveal their privacy information, such as age, identification, address, physical condition. For the above issues, this paper proposes a privacy preserving target pattern matching scheme (PP-TPMS). Our scheme utilizes bloom filters and secret sharing technology to realize secure pattern matching between given query requests and mass medical information. The auxiliary diagnosis results are returned to users. The experimental results show that this scheme has efficient computation performance and communication performance.

To address the existing problems of Chinese named entity recognition, the traditional pre-training model fails to characterize the multiple meanings of words and the model does not sufficiently dig into the potential semantic features at the Chinese word level. This paper proposes a Chinese named entity recognition method based on BERT and fused attention mechanism. First, the word vector features are obtained by the pre-training of large-scale corpus with the use of BERT model to deal with the problem of multiple meanings in one word. Then, the contextual features are recognised through the use of BiLSTM and thus passing the results into the attention layer. This is to exploit the potential semantic features within the text in order to face the shortcomings of unpromising relevance with the given information of semantic feature in previous models. Last, the output results are annotated in sequential order by CRF to reduce the probability of incorrect labelling. Through comparative experiments, the F1 values of this paper's model are 95.12% and 95.43% on MSRA corpus and People's Daily corpus datasets, respectively, which are both better than the comparison models, revealing the effective improvement in the named entity recognition.

Smart education is a new realm of educational informatization, and the smart education model is an important part of the teaching ecosystem. Therefore, this paper conducts a visual analysis of the current problems in intelligence education through 422 papers from CSSCI with source journals as the core in CNKI in the past decade, and puts forward the view of strengthening the combination of artifacts and people, and the combination of theoretical research and teaching experiment. Suggestions are given to increase the degree of cooperation between authors, and strengthen the cooperative research of universities and researchers. This will sort out the research status of smart education in China and the development status of smart education, and provide meaningful reference for the future development of smart education.

The goal of this paper is to implement the ORB-SLAM3 algorithm on underwater vehicles and evaluate its performance in the underwater environment. We tested the ORB-SLAM3 algorithm using the AQUALOC dataset, which is recorded in different underwater environments close to the seabed by ROVs. Additionally, we built an experimental platform based on the BlueROV2 equipped with a mono-camera and an inertial measurement unit (IMU). We calibrated the mono-camera based on a pinhole camera model and ran the ORB-SLAM3 algorithm with limited perceptional possibilities BlueROV2’s sensors provide. The experiment shows that the visual-only module of the ORB-SLAM3 can effectively locate the underwater vehicle in structured areas, and its visual-inertial module can locate the ROV well even in unstructured areas.

Unmanned aerial vehicle (UAV) and intelligent reflecting surface (IRS) are two promising technologies which can improve the signal coverage and network throughput. However, different practical restrictions of these two technologies restrict their applications. In this paper, the maximum sum-rate is investigated in the orthogonal frequency division multiple access (OFDMA) system aided by IRS-equipped UAV. We comprehensively optimize the bandwidth allocation, the flight trajectory and the phase shift. The problem is non-convex that is difficult to solve directly. Therefore, we break down the problem into several sub-problems, and develop an alternating optimization (AO) algorithm to gradually obtain the suboptimal solution. After obtaining the BS resource block allocation solution, the successive convex approximation (SCA) technique is used to obtain the trajectory and the phase shift solution. Simulation results show that the simulation results show that the UAV-assisted OFDMA communication system equipped with IRS can greatly improve the system and speed.

The internet of vehicles industry is a new industrial form of automobiles, electronics, information and communications, and road transportation. It is a hot spot for global innovation and a future development trend. As the country attaches importance to data security, the internet of Vehicles, as the direction of future auto industry transformation and upgrading, is also facing huge security challenges. Among them, data security is the focus of current internet of vehicles security research and management. This article first focuses on the threats of data leakage, data theft, and data tampering faced by internet of vehicles data, describes the data security risks. Analyze the disadvantages of the current solutions for data security in the internet of vehicles. Then proposed the use of blockchain technology to protect data. Finally, relevant suggestions were put forward in terms of building a firm bottom line of data security, advancing the construction of a data security system for the internet of vehicles, strengthening basic technology research and blockchain-based big data protection security mechanisms.

In order to meet the requirements of the navigation status monitoring of unmanned surface vehicle (USV), a navigation status monitoring system of (unmanned surface vehicle) is designed and implemented. The system uses LoRa technology to realize the data interaction between USV and the shore-base terminal, and the shore-base terminal analyzes, processes and stores the return data to realize data visualization, track query, historical data query and other functions. The test result show that the system can meet the requirements of real-time monitoring of the navigation system of USV, and provide reference for the related research of USV.

This study introduces a graphical user interface (GUI) based on MATLAB to realize the automatic extraction of sizes of defects from the infrared sequence. To obtain the edge of the defect at deeper layer, a fusion stratagem of the maximum of gray values is adopted for an image subset in the sequence. Blob analysis to the fusion image is used to obtain the general information of defects of a specimen including the distributions and numbers of defects. The frame image is determined for a certain defect according to the peak of the time history curve of sensitive region variance. It can yield a region of interest (ROI) to expand the blob in the selected frame and the defect can be acquired by image segmentation. The results show that through this GUI, a better thermal image can be selected from a set of infrared sequence diagrams for quantitative extraction of different buried depth defect areas, which realizes automatic defect extraction, and its relative error is within 5%. The research on infrared automatic detection technology has certain significance.

A novel authenticated encryption chip for radio frequency communication application was developed. A range of low power techniques were applied to reduce the power consumption. The proposed authenticated encryption chip exhibits ultra low power dissipation of 139.2 μW/MHz @ 1.8V V_DD and ultra high throughput of 1.07 Gbps with a reasonable core area of 0.461 mm2 (46.8 k gates) under SMIC 180 nm technology.

The existing high frequency (HF) network frequency allocation uses fixed frequency point set, which is difficult to meet the demand of HF cognitive network dynamic spectrum access. According to the characteristics of HF frequency point fading, a HF frequency allocation method based on combination auction is proposed for HF cognitive network. Based on shortwave cognitive network characteristics, based on the short wave shortwave network, the paper designs combinatorial auction process of cognitive radio technology, builds the mathematical model of combinatorial auction to reduce time selective fading effect on the efficiency of the auction algorithm, introduced the gated recurrent unit (GRU) helped forecast model, realizing the frequency point during the week in predicting, improve the average buyer yields, increasing the efficiency of the frequency spectrum assignment. The simulation results show that the performance of the combinatorial auction algorithm increases with the improvement of spectrum sensing ability, and the average buyer's return rate increases significantly in the fading condition through frequency point state prediction. When the number of combinatorial frequency points increases, the link reliability is enhanced, but the average buyer's return rate decreases.

The images generated by the existing makeup transfer methods have some problems, such as large loss of facial structure and inconsistent color distribution with reference images. In this paper, a makeup transfer method is proposed, which can keep facial structure information unchanged. Firstly, this paper builds an efficient generator model by using the characteristics of U-Net which combines up-sampling and down-sampling information to extract image features and SE-Net which emphasizes useful features and suppresses useless features. At the same time, a loss function is designed to constrain the facial color distribution of the generated image so that the generated image is as consistent as possible with the color distribution of the reference image. Experiments show that the makeup transfer method in this paper not only can capture the color distribution of the reference image's face, but also better preserves the facial features of target image with an average SSIM of 0.8740.

As an important platform to support cyberspace security technology verification, cyberspace attack and defense equipment performance testing, and cyberspace attack and defense rehearsal confrontation; the network security test platform plays an increasingly important role in the field of network security research. With the rapid development of cloud computing and big data, the network security test cloud platform shows the characteristics of explosive growth of data volume, among which the value of massive log data is particularly important. Through log analysis, abnormal events and behaviors can be found in a timely manner, but the traditional log detection technology appears to be incompetent for the analysis of massive log data, and the log detection and analysis technology based on Elastic Stack can realize real-time collection and retrieval of massive log data, and then realize effective detection and analysis of abnormal events in the network security test cloud platform.

The rapid development of E-health system brings more convenience and flexibility to patient. However, it also faces several challenges. The private information which transmitted in the E-health will be used for medical diagnosis, so during transmission the issues of data privacy are the most concerning for users of E-health systems. In order to protect the private information of patients, we present a three-factor authentication scheme for E-health system. Furthermore, fuzzy extractor is added to our scheme, which can protect the biometric information of patient. Finally, analyzing the security and performance present usability of our scheme.

Aiming at the problems of high risk and low real-time performance of concrete temperature rising in monitoring concrete temperature in buildings, a real-time monitoring system based on embedded high precision concrete temperature was designed. The system realizes the functions of real-time temperature detection, real-time instruction and remote monitoring. The system uses MQTT protocol to communicate between client and server. It works on TCP/IP protocol family and provides orderly, lossless and bidirectional connection. The device then sends data in a custom format through a custom topic class, and the Internet of Things platform stores and processes the flow of device data. In addition, users can import log files to MATLAB and other software for subsequent analysis according to their needs. Through the concrete temperature change curve and cooling rate curve marked with latitude and longitude, abnormal temperature measurement points and problems in the construction can be quickly located, which is convenient for timely adjustment of the construction scheme system. This paper designed an intelligent concrete temperature monitoring system with high precision, high reliability and high real-time performance.

In this paper, we compared the differences of acoustic characteristics between synthetic speech and emotional speech under the same text from the perspective of the lack of emotional expression of synthetic speech by using Praat software for a single phoneme /ei/. Analyzing the results, it was concluded that the differences of the emotional information were mainly in the small dispersion of synthetic speech fundamental frequency, the dispersion of synthetic speech intensity was much smaller than that of real speech with large emotional fluctuations, the harmonic waves in narrowband spectrograms were nearly straight without bending and jittering, the formant center frequencies interlacing degree is small. The common differences between synthetic speech and neutral and emotional speech were shown by the absence of harmonic waves at frequencies above 3000 Hz and the obvious difference in the direction of the tail end of the second formant.

Artificial intelligence technology has a far-reaching impact on the way of life and production of human society. It not only deeply integrates with various industries, but also promotes the development of industries. In the era of comprehensively promoting the concept of green development, will artificial intelligence technology promote green innovation? Based on this background, this paper first constructs a theoretical model of the relationship between artificial intelligence technology and green innovation, and then makes an empirical test with data. The results show that AI technology can significantly promote the development of green innovation and enhance the positive effect of government subsidies on green innovation.

Blood pressure is an important data about our health. Accurately measuring blood pressure and controlling blood pressure within the normal range is of great significance to our health. Know whether your blood pressure is in the normal range at any time to reduce the incidence of blood pressure diseases. According to the pulse wave theory and oscillographic method, a blood pressure measurement system is designed by amplifying small signals and detecting and processing technology. The basic principle is: the blood pressure sensor detects the pressure value in the air bag, and the air pressure in the air bag decreases slowly. At this time, the air pressure sensor converts the collected human pulse wave signal into a voltage signal, which is amplified and filtered, and then transmitted to the system for data processing. Finally, the blood pressure is displayed and the measurement is completed.

This paper proposes a power equipment indicator status detection algorithm based on the improved YOLOv4 and HSV color space. Firstly, the image of power meter is preprocessed by denoising and histogram equalization. Secondly, the PSA attention mechanism module is added to the network layer of YOLOv4 to improve its power of small target detection. The improved YOLOv4 is used to detect the indicator lights in the image, the color of the detected indicator light is determined through the HSV space. Experimental results verify that the proposed algorithm has better performance than several existed algorithms.

Affective computing is an interdisciplinary research area that includes machine learning and pattern recognition, psychology, and cognitive science. The aim is to research and develop theories, methods and systems that can recognize, interpret, process and simulate human emotions. In this article we propose a neural network model for multimodal emotion recognition based on cross-media data-feature fusion. Multimodal data fusion can effectively improve the accuracy of emotion recognition. We extract features from EEG data and facial images using a deep double-stream neural network and then merge them in a medium-term feature layer to identify three categories of emotions (sadness, calm, and happiness). The experimental results show that the detection accuracy can reach over 95%. Compared to the traditional single-modal emotion recognition method, the accuracy rate of emotion recognition based on EEG data and facial images has been significantly improved. It also proves that the multimodal medium-term feature layer fusion method has good applicability for emotion recognition.

Face alignment and face reconstruction are hotspots in computer vision and artificial intelligence. Aiming at the problem of large model and low accuracy of large pose face alignment algorithm, a new lightweight network model FTCNet is designed and implemented. First, deep separable convolution is used to build a lightweight deep neural network model, which directly inputs images in an end-to-end manner for face alignment. Secondly, the network model classifies and predicts pose parameters, shape parameters and expression parameters, and improves the performance of the model through the feature transfer mechanism. Tested on multiple datasets, the experimental results show that the FTCNet gets small, fast, and high-quality face alignment and reconstruction results for unconstrained face images.

In order to improve the matching rate of weak and repeated texture regions, caused by a wrong match in somewhere of the image, which will be propagated to adjacent pixels for the non-local aggregation algorithm based on minimum spanning tree image segmentation, a segmented regions collaborative optimization for stereo matching algorithm based on segmenttree (ST) is proposed. In the cost calculation stage, the cost calculation method of multi-feature fusion based on color and gradient is used from the perspective of probability to obtain the initial matching cost. Then, in the cost aggregation stage, the segmented regions collaborative optimization algorithm is used to filter the segmented regions based on ST, and the collaborative optimization strategy of matching cost within the region is realized by selecting the main disparity and the proportion-based penalty optimization method. Finally, the disparity map is obtained through some post-processing. The algorithm test on Middlebury’s V2 datasets. The experimental result shows that the average error matching rate of this algorithm is 5.72%, and it shows that the algorithm in this paper has a high matching accuracy.

In recent years, with the continuous expansion of the business scope of auto insurance, the crime of auto insurance fraud is becoming frequent. The establishment of auto insurance fraud detection model has become an important measure to ensure the stable development of auto insurance industry. This paper builds an auto insurance fraud detection model based on Logistic-SVM, which could solve the problem that the original model needs lots of variables. Firstly, the importance of characteristic variables is sorted by SVM model, and ten characteristic variables are selected according to the objective reality. By comparing with the traditional single logistic regression and SVM algorithm, it is found that the Logistic-SVM algorithm has a better detection effect on auto insurance fraud. The accuracy of Logistic-SVM is 96.1%, which is 2% higher than that of logistic-regression and 0.7% higher than that of SVM. The research of this paper could not only improve the practicability of machine learning model in the field of auto insurance fraud detection, but also escort the prosperity and development of auto insurance industry.

The genetic algorithm is an intelligent optimization algorithm derived from the law of biological evolution. Due to the weak local optimization ability of the algorithm and the increase in the number of iterations, it is easy to cause the lack of diversity of the entire population to affect the search effect. The Cuckoo search algorithm uses the Levy flight to update the position and status. The Levy flight can help the Cuckoo search algorithm avoid the shortcomings of local optimality and insufficient population diversity in the optimization process, so this paper proposes an improvement that integrates Levy operator genetic algorithm. The test function is used to test the improved genetic algorithm optimization ability and robustness and compare it with the genetic algorithm, Cuckoo search algorithm, and particle swarm optimization algorithm. The simulation results show that the algorithm has good optimization ability and robustness, and the algorithm is significantly better than the original genetic algorithm in optimization accuracy and robustness.

The rapid development of the global economy has brought a lot of fossil energy consumption and environmental pollution, such as the greenhouse effect caused by car exhaust. In order to fundamentally replace the use of fossil energy, electric vehicles have been vigorously promoted by governments all over the world in recent years. However, the electric vehicle charging pile has encountered a new problem in the process of promotion: the electric vehicle charging load is often unbalanced in time and space, which requires an accurate power load forecasting and scheduling model. In the past, algorithms such as random forest were used to predict the load data of charging piles, which provides a more accurate prediction for the load data. However, these methods require a large amount of data trained by the power load model and are not conducive to the protection of privacy. In order to solve these problems, we design an FRF-CNN model, which combines federated learning with random forest and the convolutional neural network model. Extensive experiments show that FRF-CNN has better classification performance on distributed charging piles than other models, and our method effectively protects the privacy of sensitive data.

In recent years, the auction industry has developed rapidly, and online auctions have become increasingly popular. However, the development of online auctions has also brought risks such as Shill bidding. This paper builds a Shill bidding prediction model based on support vector machine algorithm to solve the problem of difficulty in predicting Shill bidding behavior. Through the sorting and analysis of the characteristic data in the Shill bidding cases, ten indicators that are significantly related to the Shill bidding behavior have been obtained. In order to overcome the imbalance problem of the training set, a sampling balance mechanism is introduced to sample the data set. By comparing the calculation results of logistic regression and naïve Bayes algorithm, it is found that the support vector machine algorithm has the highest accuracy of Shill bidding risk prediction, reaching more than 99.2%. This study could not only improve the auction industry's ability to monitor, analyse, and judge the early warning, monitoring, analysis and judgment of bidding behavior. It could also guarantee the healthy and sound development of bidding work, and play a role in escorting social and economic development.

Limited by the difficulty of vehicle speed control when registering QR codes and the high maintenance cost of QR codes, the development of automated guided vehicle (AGV) navigation technology based on QR codes has encountered bottlenecks in the development of industrial logistics. The information fusion of the magnetic positioning method and the odometer can overcome the shortcomings of the AGV navigation technology based on the two-dimensional code. By laying magnetic nails on the ground, the magnetic positioning method provides the attitude of the closest magnetic nail relative to the automated guided vehicle. The experimental results show that the average path accuracy of the magnetic nail navigation based on the EKF algorithm proposed in this paper is 83±15 mm, and the average magnetic positioning accuracy is 11.7±0.59 mm.

Aiming at the path planning problem of logistics unmanned aerial vehicles (UAVs) that can deliver multiple packages, this paper comprehensively considers the maximum travel distance of UAVs, the maximum carrying capacity, the geographical location of delivery centers and delivery points, and the priority of packages, with the shortest flight distance and the highest customer satisfaction as the objectives. A multi-objective programming model with multiple constraints is constructed and an improved simulated annealing algorithm is designed. By constructing a specific solution space and calculating the probability of accepting new solutions, the global optimal solution can be obtained by avoiding local optima as much as possible. Finally, the sensitivity of initial temperature, temperature gradient, and iteration times of simulated annealing is analyzed. The results show that the simulated annealing algorithm can efficiently plan the UAV navigation path and minimize the cost. When the initial temperature is 180℃~ 220℃, the temperature gradient is 4℃, and the number of iterations is between 200 and 300, the probability of obtaining the global optimal solution is the highest. The research results can provide guidance and reference value for the application of UAVs in logistics distribution systems.

Humor detection has recently become one of the most popular topics in NLP. Now existing humor datasets are mainly designed to analyze whether most of the text is funny. SemEval-2020 Task 7 was given to assess humor in edited news headlines. Our task is to analyze how one-word or one-phrase editing could convert a text from non-funny to funny by predicting humor ratings for edited news headlines. We replace the regression task with the three-category and two-category classification tasks. We use a pre-trained BERT model. The results suggest that humor of headlines with atomic editing are not significantly detected by our model for tasks including regression, three category classification, and two-category classification. We discuss the process of how we update our model and possible explanations of why the model is not significant. Moreover, we analyze features under each label in the dataset and data distribution in the dataset.

This paper selects the input-output data of 11 large listed power generation enterprises in China during from 2014 to 2019, and calculates the static total factor productivity and its decomposition value based on the BBC model. The results show that only the total factor productivity of C9 enterprises is DEA effective, and the total factor productivity indexes of the other ten power generation enterprises are data envelopment analysis (DEA) invalid. In addition, the decomposition results of total factor productivity show that the pure technical efficiency DEA of C1, C4, C8 and C9 enterprises is effective, and the scale efficiency DEA of C9 enterprises is effective. As the main field of energy production, each enterprise should reasonably adjust the enterprise scale and resource allocation according to its own situation to promote the high-quality development of the enterprise.

Breast cancer is currently one of the most common and fatal cancers in the world. Studies have shown that ERα is an important target for the treatment of breast cancer, and how to effectively develop compounds that can antagonize ERα activity and have a good ADMET properties is a key factor in the development of breast cancer and related drugs. Around the problem of feature reduction and screening of sample data, based on the highly nonlinearity and strong coupling between the data samples, this paper innovatively uses the hierarchical analysis method (AHP) to combine feature selection with feature dimension reduction, the correlation of pIC50 and the remaining 793 variables was studied using Pearson correlation, Spearman correlation, Kendall correlation, recursive feature elimination, and random forest. Different methods were weighted by hierarchical analysis, and the highly correlated variables were removed by the independence test. The main variables such as MDEC-23, MLogP, LipoaffinityIndex were finally obtained. Centering on the prediction and optimization of ERα as an anti-breast cancer candidate, this paper proposed three quantitative prediction models of ERα biological activity (Gaussian process regression and XGBoost). We experimentally analyzed the prediction effect of the three models, finally determined the XGBoost distributed prediction model with a goodness of fit of 91%, and introduced various evaluation indicators to verify the accuracy and scientificity of the proposed prediction model.

In view of the increasing number of pets in urban communities, the relationship between urban residents and pets in urban centers has become increasingly complex, and the entanglement between people and pets in urban centers has gradually escalated conflicts. Drawing lessons from Hong Kong’s experience in transforming urban micro-spaces into dog-walking parks, the design ideas for dog-walking parks or dog-walking areas are proposed, in order to provide a reference for the future construction of dog-walking parks in my country. Under the concept of urban micro-renewal, this paper selects a 100-square-meter space in Yuyuan Road Community, Shanghai as the public physical space reconstruction project, and uses digital modeling software, Rhino, to model the reconstruction plan in three dimensions. In the modeling process, based on the previous field investigations and questionnaires, the needs for interaction between people and pets in outdoor public spaces were summarized. The pet factor forms active partitions, activates the community, and creates a good quality of living space.

To solve the problem of low detection accuracy of steel surface defects due to background interference and various target shapes, a steel surface defect detection algorithm with attention mechanism is proposed to improve detection accuracy. In view of the small proportion of the target defect area in the overall image and background interference, a two-way attention module (TWA-Block) is proposed to establish the long-distance dependence of the spatial domain and channel domain features. It enhances the contour and texture features of defect area in shallow features, and suppresses the background to a certain extent. The experimental results show that the average accuracy (MAP) of the YOLOv3 model fused with the attention mechanism on the NEU-DET dataset reaches 79.5%, which is 14.4% higher than the YOLOv3 algorithm. Compared with the standard steel surface defect detection methods, the algorithm effectively improves the detection accuracy.

The mode recognition system is designed to be built by using computer software technology. On the one hand, it is designed to functional layer the mode recognition system through data transmission and data analysis to ensure the security of the data. On the other hand, in order to better conduct module management, modular management can make different command functions that are divided and managed, but also can make the software to get more accurate positioning. In addition, the mode recognition system built based on the computer software technology can be widely used in the electronic product-related fields, and the system can automatically recognize the voice and start the corresponding functions.

In the context of the era of big data, the emergence of e-commerce platforms has brought many opportunities and risks. Due to the COVID-19, e-commerce has achieved unprecedented development, and e-commerce fraud has severely damaged the healthy economic environment. This paper uses the RUSBoost algorithm to build an e-commerce fraud risk prediction model, and verifies the predictive performance of the model through data experiments. The results show that it has a high accuracy rate for identifying e-commerce fraud. If the model is applied to e-commerce, the losses caused by ecommerce fraud could be avoided in time. At present, there are fewer e-commerce fraud risk prediction models and have a wide development prospection.

To meet to the requirements of the national production safety law of China, large manufacturing enterprises need to implement the sampling of dynamic number of employees in key areas, emergency evacuation drill and emergency evacuation organization. Aiming at the existing problems such as high randomness of overtime work of workshop employees, frequent worker aggregation phenomena for short meetings, fast aggregation, short time and high risk of dining employees in the canteens, uneven personnel distribution and crowded entrances and exits in the conference center, this paper analyzes the psychological and moving behavior characteristics of personnel in these typical scenarios, and introduces the convolution neural network (CNN) model in the field of machine vision, to build a multi-scenario employee number statistics and density estimation model for factory workshops, canteens and large meeting centers. Further, according to the density-risk relationship, a crowd aggregation risk early warning model is established. Finally, taking the video surveillance system (VSS) as the data acquisition source, the application cases of practical scenes such as workshop, canteen and meeting center are designed to verify the effectiveness of the density estimation model and aggregation risk early warning model proposed in this paper. Thereby this paper provides technical guarantee for the safety of employees in large manufacturing enterprises.

High resolution range profile system is usually used in air-to-surface radar to obtain radial range distribution characteristics of multiple scattering centers. Because clutter exists in the scene, constant false alarm detection is needed for high resolution range image to obtain target range image. When the detection threshold is low, more false-alarm will be introduced into the detection results, which will affect the classical extended target detection effect. To solve this problem, a new method is proposed in this paper, which applies monopulse angle measurement and weighted fuzzy C-means algorithm with scattering point amplitude as weighting coefficient to constant false alarm detection algorithm with prior information of target quantity and volume. With the same detection threshold value, the location information and cluster center parameters are analyzed and judged simultaneously, which eliminate effectively false-alarm and obtain accurate target range image. This method has been verified by simulation results.

Aiming at the actual wireless sensor network (WSN), this paper studies the location of all sensor nodes in WSN. How to efficiently route multiple mobile chargers to minimize the energy consumption of all mobile chargers and minimize the battery capacity of all sensors. Firstly, according to the coordinates of all sensor nodes, the corresponding sensor network is determined, and the constraint conditions and objective function are determined according to the actual sensor network, and then the mathematical optimization model is established. This problem can be analogous to the VRPTW problem (VRP with Time Windows), namely a vehicle path planning problem with Time Windows. The shortest loop path of multiple mobile chargers can be obtained by multi-objective optimization and improved genetic algorithm. Then, the minimum capacitance of each sensor is determined according to the sensor nodes contained in the loop of each mobile charger. Finally, the improved algorithm has a good convergence degree and the model has good practicability through computer simulation. This study can provide a good theoretical reference for a wireless sensor network in the configuration and mobile route of mobile chargers.

Multifunctional radar (MFR) has become an indispensable part of cognitive electronic reconnaissance due to its strong electromagnetic sensing ability and high adaptability. Most of the existing researches focus on the individual identification and work pattern recognition of MFR. Although it can provide certain intelligence support, it cannot serve cognitive electronic warfare intuitively. Based on the theory of behavioral science and based on the working characteristics of MFR in the context of cognitive electronic warfare, this paper innovatively proposes a behavioral modeling method for MFR, which lays a good foundation for subsequent behavior recognition and behavior prediction.

In the improvement of many meta-heuristic algorithms, two or more algorithms are often mixed into one algorithm to improve the convergence speed and accuracy of the algorithm. The search ability of Harris Hawk optimization (HHO) algorithm ^[1] is relatively weak, and the search ability of Aquila Optimizer (AO) is relatively strong. So, in this article, the Aquila Optimizer (AO) algorithm is mixed on the basis of the piecewise linear map enabled Harris Hawk optimization (HHO) algorithm, and the advantages of the AO algorithm are used to complement the disadvantages of the piecewise linear map enabled HHO algorithm. This hybrid algorithm is called CHHOAO. The results show that this hybrid improvement is effective and can significantly improve the optimization ability of the algorithm.

Image registration for SAR lays the foundation for interferometric synthetic radar. And its image registration quality directly affects the extraction accuracy of surface deformation. Owning to the optimal performance of PS-InSAR for monitoring surface deformation in urban areas, the article adopts PS-InSAR to monitor the surface deformation in the study area. PS points with accumulated deformation range from -3.568 to 4.680 mm. These points are selected as ground control points (GCPs) for image registration for SAR. In that case, it is possible to make some improvements on traditional registration algorithm. By means of filtering processing for both the improved algorithm and interference pattern guided by traditional algorithm through Boxcar, Goldstein, Adaptive and non local InSAR, the percentage of areas with coherence over 0.6 can be calculated. In the final place, it is discovered that the improved algorithm has a significant impact on enhancing all four filtering algorithms. Among these filtering algorithms, the performance of non local InSAR is improved most significantly, with an improved value of 8.81%.

Whale optimization algorithm (WOA) is characterized by fewer parameters, simple structure, and stronger optimization seeking ability compared with traditional optimization algorithms, but in practical applications there are problems such as sluggish convergence speed and easily falling into local optimal solutions. This work proposes MAWOA, a whale optimization algorithm based on hybrid adaptive strategy, introducing a method to adaptively adjust the weights with the iterative situation of the population to accelerate the convergence of the algorithm; designing an adaptive adjustment threshold, and individuals select a random search method according to the value of the threshold to enhance the global search ability of the population and circumventing local values; introducing an adaptive nonlinear convergence factor to strengthen the algorithm in initial exploration breadth and later local development process. Twelve different morphological benchmark functions and a MAWOA-BP wine quality classification model were used to optimize the experiment. The results shows that MAWOA has stronger performance in terms of convergence speed and optimization-seeking accuracy, and the classification results are significantly improved compared with traditional classification models such as KNN, decision trees and BP neural networks.

Currently, for the problem of personal credit risk identification, the most commonly used method is to optimize the parameters of the model through bionic algorithms to obtain higher accuracy, but it may face the risk of lower precision. Some scholars also discussed the identification of personal credit risk from the perspective of combination models. From the perspective of integrated learning, based on C5.0 algorithm and using boosting technology, this paper constructs the boosting-c5.0 personal credit risk identification model, and uses UCI German personal credit data set to verify the performance of the model. The study found that the accuracy, recall, precision and AUC value of boosting-C5.0 model are better than SVM, logistic and C5.0 models.

Regarding the current existing modeling methods, procedural modeling is a fast and convenient method. The more perfect shape in use now is the CGA shape, which is a new shape syntax for procedural modeling of CG architecture. It can clearly show the outline and appearance of the building, but the process is too cumbersome and the algorithm is difficult. Based on this problem, this paper proposes a method of programmatic generation of building models based on triangulation algorithm and UV texture mapping. This article mainly constructs a procedural modeling method for a single building.

Footprints are important information at the crime scene and play an important role in the field of criminal investigation. At present, the research on footprints mainly focuses on barefoot footprints, but the main thing obtained at the crime scene is shoeprints. How to mine barefoot footprints through shoeprints is one point of the key problems in the field of footprint recognition. This paper takes optical footprints as the research object, collects 95 people’s cloth shoeprints and barefoot footprints, and proposes a generative adversarial network which combines self-attention modules and multiscale discriminator (SM-GAN). The self-attention module is added to the generator, which enables the network to focus on the association between footprint structures. The discriminator uses a multiscale discriminator structure, which improves the generation effect of the generated image in the global and local areas. The experimental results show that the method proposed in this paper has a better effect of generating footprints than traditional image-to-image translation methods.

At present, with the rapid development of science and technology such as computer and electronic information, it also drives the renewal of technology application in emergency prevention and control. Information fusion is a new technology developed to meet this demand. Image fusion technology is an important branch of multi-sensor information fusion. By analyzing the current situation and problems of emergency prevention and control, this paper puts forward the necessity of using image fusion technology, introduces the current situation, levels methods and application scenarios of image fusion technology, and explores the application of image fusion monitoring in emergency prevention and control. It is of great significance for the prevention and control of emergency events and the construction of social security.

The technology of vital signs detection for the heart rate measurement using a Doppler radar system has been proven of great use, whereas it is still limited by several challenges. The major challenges are respiratory harmonic interference and random body movement (RBM), which significantly degrade the accuracy of measurement. In this paper, a novel vital signs detection method is proposed to acquire accurate heart rate measurement by jointly exploiting the respiratory harmonic cancellation method and regional hidden Markov model (RHMM) technique. The complex signal demodulation technique is firstly used to acquire an accurate respiratory rate estimation, and the respiration harmonic cancellation method is then introduced to address the harmonic interference. Finally, the RHMM technique is utilized to acquire robust and enhanced heart rate measurement by exploiting the underlying slowly-varying characteristics of the heartbeat signal to mitigate the effects of RBM. Experimental results based on a 24 GHz Doppler radar system show that the proposed method has a better performance than previous methods, whose average accuracy of the heart rate measurement reaches up to 94.5% in the RBM environment.

5G network slicing with low delay and high reliability has been greatly sought after in the intelligent network connection automobile industry, and the development of various scenarios of vehicle networking also reflects the gradual deepening of 5G to various fields. Therefore, the embedded connected and autonomous vehicle communication terminal is analyzed and designed according to the application scenario of the intelligent networked automobile, and the Monte Carlo algorithm is utilized to maximize the vehicle data throughput and the total system throughput of the end-users by constructing the 5G mobile communication network slice model. The simulation results suggest that under the optimized power allocation technology, the end-users who receive vehicle data are enabled to achieve high throughput in their slices.

With the development of passive detection system, aircraft communication is being seriously threatened. In order to improve the stability of aircraft network cooperative mission communication, a routing protocol design method of aircraft formation clustering communication based on evolutionary game theory was proposed. This method constructs a dynamic networking protocol based on ad-hoc, which realizes the selection Pareto equilibrium in the network cluster by using the evolutionary game theory. This paper puts forward the concept of life factor, which solves the problem of network instability caused by frequent disconnection of communication links, and finally realizes the combination of evolutionary game theory and the communication design method of routing protocol in aircraft formation cluster. Compared with the traditional routing protocol, the aircraft formation clustering communication routing protocol based on evolutionary game has the best survival rate in the mission cycle. In the same mission scenario and under the same conditions, the normalized life factor decline rate of the evolutionary game clustering communication routing protocol is significantly lower, and has better RF stealth performance.

According to the periodic modulation characteristics of UWB signal, its spectrum is discrete. Theoretically, only at a specific frequency point, the spectrum amplitude is non-zero. Using this feature, under the condition of equivalent sampling, according to different sampling rates, the appropriate spectrum amplitude is selected in the spectrum of the sampled signal to remove the noise and interference components, and then the time-domain signal is interpolated and low-pass filtered through the inverse Fourier transform to return the original signal and improve the signal to interference ratio of the signal.

In recent years, as a key support technology of big data, the software-defined network has been widely studied as a new network innovation architecture. This article introduces the technology of the space-ground integrated network and software-defined network in detail. It discusses the research progress of software-defined network (SDN) in the architecture of the space-ground integrated network and compares the advantages and disadvantages of single-layer controller deployment and multilayer controller deployment. Finally, through the analysis of the characteristics of the space-ground integrated network, consider the GEO satellite deploy master controller, part of the LEO satellite deploy from the controller, and multiple performance indicators for controller deployment issues, the controller deployment issues, through a variety of optimization algorithm of satellite number more LEO satellite experiment simulation controller.

In order to meet the data requirements in the development of space-based optical observation platform and improve the simulation accuracy of space debris tailing, a simulation method of space-based optical observation image based on two-line-elements is proposed. Firstly, by analyzing the imaging process of real image, the simulation process of space-based optical observation image is designed, and the calculation method of each process is given. Then, the space debris tailing of real images is analyzed, and the limitations of the traditional space debris tailing simulation algorithm are obtained. Finally, combined with two-line-elements, a space debris tailing simulation algorithm suitable for long exposure time is proposed. The simulation results show that the algorithm not only ensures the fidelity of imaging, but also simulates the tailing of space debris with high precision, which is of great significance to the development of space-based optical observation platform.

With its large transmission capacity, low communication loss, strong anti-interference ability and long transmission distance, optical cable has become an important transmission channel of longitudinal differential protection information. Whether it is direct route or circuitous route communication mode, the interfaces and cables of protection devices and communication devices may fail, and these devices and cables are maintained by different operation and maintenance departments. When a fault occurs, the mutual cooperation of multi department operation and maintenance system and operation and maintenance personnel is required to locate the fault. The efficiency of troubleshooting is low, which seriously reduces the power supply quality of the power grid. By introducing semantic technology, this paper correlates the data attributes between various systems, and finally locates the location information of faulty equipment from the initial PMS alarm, so as to realize the intelligent fault search, which provides convenience for the operation and maintenance personnel to locate the fault.

The electromagnetic exposure from mobile phones has attracted the attention of many researchers. In this paper, we established four kinds of mobile phone simulation models according to the different antennas and phone shells. These models also include the mobile phone components of protective glass, conductive glass, screen back plate, battery, and PCB dielectric slab. We studied the effect of the shell material on the specific absorption rate (SAR) of energy deposition in the body, such as plastic and metal shells. We compared the exposure levels of the mobile phone with the planar inverted-F antenna (PIFA) and the planar monopole antenna (PMA). We analyzed the SAR distribution when a person is in sitting or standing postures, or holds the top, middle and bottom of the mobile phone.

At present, the analysis and identification of unknown targets mostly rely on manual labor, which can not make fast and accurate decisions for the current situation. As a knowledge representation method in the era of big data, knowledge graph has the capabilities of semantic processing, open interconnection, and logical combing. In this paper, we construct a knowledge graph of different platforms with radar through three main steps: data acquisition, knowledge extraction, and knowledge storage, and then propose a knowledge reasoning framework. Based on it, when faced with massive data, combined with the obtained related intelligence information, we can effectively analyze and identify the type of target platform, and then estimate the opponent’s actions and intentions, and assist us to make command decisions.

The digital filter is widely used in many situations with the development of digital signal process techniques. The basic theory of digital filter is briefly described at first. Then the common design methods of FIR are introduced, and different window functions are demonstrated in detail. And different design method of IIR methods are compared and analyzed. For a given experimental example, the required FIR and IIR low-pass digital filters are designed respectively by MATLAB program function and FDAtool. Finally, the simulation models are constructed in Simulink, and the results show that both designed FIR low-pass digital filter and IIR low-pass digital filter can filter the mixed signals to obtain the required 1 MHz sine wave.

A novel 3300V conductivity modulation enhanced planar gate IGBT (CE-IGBT) with P-type Schottky contact and partial N-type buried layer is proposed. The proposed CE-IGBT features a P-type Schottky contact between the P-base region and emitter metal as well as a partial highly-doped N-type buried layer located in the P-base region. The Schottky contact forms a hole barrier by increasing the potential of the P-base region and the N-type buried layer suppresses the hole flowing above it. The simulation results show that compared with the conventional planar gate IGBT (Con-IGBT), the proposed CE-IGBT not only reduces the on-state voltage drop (V_ceon) but also improves the trade-off relationship between the V_ceon and turn-off power loss (E_off). Compared with the Con-IGBT, the V_ceon of the CE-IGBT with same P-base doping concentration is reduced by 32.7%, and the CE-IGBT pro with same threshold voltage (V_th) is reduced by 30.9%. At same V_ceon of 2.65 V, compared with the Con-IGBT, the E_off of the CE-IGBT and CE-IGBT pro is reduced by 37.3% and 34.9%, respectively. Moreover, the proposed device demonstrates good reverse biased safe operating area (RBSOA).

It is the key step of automatic apple picking and sorting to segment of the apple image in natural environment using image information processing technology. In order to improve the intellectualization of apple picking process and effectively extract the target of image segmentation, a novel image segmentation method of Huaniu apple based on pulse coupled neural network (PCNN) and watershed algorithm is proposed. Firstly, the PCNN model is improved and optimized on the basis of image RGB to L*A*B space conversion. Then, the improved PCNN combined with watershed algorithm was used to segment the apple image on considering the image texture and color feature information. The theoretical analysis and experimental results show that the novel method is effectively able to segment target image, and can preserve the image details and the edges as well. The proposed algorithm has better subjective vision effect and objective quality than the traditional watershed and other related algorithms.

Over-voltage faults in the distribution grid need to respond and deal with power equipment in time due to their great damage to power equipment and systems. When the power system realizes the ubiquitous Internet of things, over-voltage faults do not have to enter the ubiquitous power Internet of things cloud platform for data acquisition and integration, but edge computing can be conducted by localized control and distributed processing, data transmission and communication are an important basis for realizing edge computing. For the ubiquitous Internet of electric power, the application framework of 5G communication technology in over-voltage fault edge computing is proposed, the distribution grid fault identification and response model based on edge computing is built, and we imagine 5G communication application scenarios.

Low power wide area network (LPWAN) is an innovative communication technology for Internet of Things applications. Combined with the overall service system and communication architecture of LPWAN, this paper analyzes the characteristics of intelligent power distribution communication network, queuing theory knowledge and service quality requirements of power distribution service, and proposes a bandwidth prediction model based on M/M/1/N queuing theory. It provides theoretical support for the application of low-power wide-area Internet of Things communication technology in distribution network service monitoring.

Unlike phased array, frequency diverse array (FDA) adopts linearly increased frequency offset across the array elements, resulting in range-angle-dependent beampattern. Thus, it has application potential in the aspect of mainlobe interference suppression. In this paper, based on the multiple-input multiple-output (MIMO) scheme with multiple matched filters at receiver, a two dimensional linearly constrained minimum-variance (TD-LCMV) adaptive beamforming approach is presented for range-angle two dimensional interference suppression. Furthermore, the anti-jamming performance is analysed in two cases, i.e., linear and nonlinear frequency offsets. Simulations indicate that the FDA-MIMO with nonlinear frequency offset outperforms that with linear frequency offset in respect of effective anti-jamming performance.

With the development of recent years, video image acquisition technology is also in constant progress and development. The technology can be widely used in all walks of life. With the promotion and development of technology, image technology for real-time and image computing requirements are constantly improving. Image storage needs logic resources also increased significantly. Aiming at this problem, it needs to develop a kind of applicable image acquisition and display system on the hardware. In order to reduce the use of logic resources and add some image processing algorithms on the system platform, this paper usesNEXYS4DDR series FPGA of Xilinx company to realize the system. The front-end image acquisition uses CMOS image sensor OV7670. The cache adopts the IP core SRAM integrated with Xilinx vivado platform. The system displays the real-time video image data through VGA display, and uses hardware description language Verilog HDL. Through simulation test, the system can fully realize the function of real-time image acquisition and display.

Aiming at the problem of unbalanced assembly load in the existing substation power line carrier smart air switch, which results in a low balance rate, a substation power line carrier smart air switch based on carrier communication technology is designed. The template matching method is used to adjust the connection piece positioning, match the standard template attributes, use carrier communication technology to improve the assembly load of the production line, plan the substation network layout, and design the substation power line carrier smart air switch according to the performance of the channel transmission device. The experimental results show that the average balance rate of the substation power line carrier smart air switch and the other two substation power line carrier smart air switches are 51.857%, 34.178%, 33.508%, which proves the substation power line carrier smart integrated with carrier communication technology. The air switch can effectively balance the assembly load of the production line and has good performance.

The design of an efficient and compact convolutional neural network (CNN) for image classification is a very challenging problem, and its design process relies heavily on the experience of experts and continuous trial and error. This paper proposed a search method based on variable scale convolutional neural network by comprehensively considering the convolutional layer, pooling layer, fully-connection layer, and activation function in CNN. In this method, a search space is designed, and the categories of search parameters are maximized. At the same time, a particle swarm variable length coding mapping method is proposed to solve the problem of coding redundancy of candidate networks. In order to effectively evaluate the performance of candidate networks, the evaluation method of the random dataset is adopted to reduce the evaluation time of CNN and improve the stability of training. The proposed search model is compared with many empirical design models and other search methods in four image classification tasks. The experimental results show that the proposed method has strong advantages and competitiveness in classification accuracy and model size compared with the existing methods.

The effectiveness of CycleGAN is demonstrated to outperform recent approaches for semi-supervised semantic segmentation on public segmentation benchmarks for a small number of the labelled data. However CycleGAN tends to generate same semantic segmentation results for acoustic image datasets, and can’t retain target details. To solve this problem, an spectral normalized CycleGAN network (SNCycleGAN) is presented, which applies spectral normalization to both generators and discriminators to stabilize the training of GANs. The experimental results demonstrate that semi-supervised training of SNCycleGAN helps to achieve reasonably accurate sonar targets segmentation from limited labelled data without using transfer learning, and surpass supervised training in detail preservation.

Taking the laser saturation interference thermal imaging camera as the background, comprehensively consider the interference distance, the rapid movement of the interference object, and the atmospheric transmission. Focusing on the problem of laser saturation interference spot position, under the external field conditions, the horizontal spot position control experiment at a distance of 1 km and the circumferential spot position control experiment at a distance of 1.7 km were carried out respectively, and the experimental data was analyzed through MATLAB. Through data fitting, the corresponding relationship between the amount of interference spot movement and the control angle is derived. When the interference angle is small, the change law of the two is close to linear.

Matching the UAV image with the airborne reference image containing geographic information can achieve the precise positioning of the UAV image target. However, there are significant differences in the imaging mechanism, image perspective and scale between UAV images and satellite maps. In view of the above reasons that lead to low image matching accuracy and large positioning error, this paper realizes an image matching algorithm based on deep convolution feature. Specifically, the multi-scale feature descriptor is constructed by using the feature maps output by different layers of convolution network, and then the feature point matching is realized based on a dynamic interior point selection method. The five shooting perspectives of the same target image of UAV are divided into vertical reference image and 45° inclination angle for matching experiments, and the performance differences between the proposed algorithm and the traditional method are compared. Experiments show that it has good positioning accuracy and better anti-angle change ability than traditional methods.

With the rapid development of mobile Internet technology and the continuous improvement of people's understanding of data security, the research on the secure transmission of information in wireless communication systems has received extensive attention. Using physical layer security technology to protect the information transmission security is not only beneficial to prevent illegal eavesdroppers from obtaining the transmitted information, but also greatly reduces the computational complexity caused by traditional encryption methods. Based on Shannon's communication theory and Wyner's wiretapping channel, this paper briefly introduces the physical layer transmission security technologies such as beamforming, artificial noise and cooperative communication, mainly analyzes massive MIMO antennas and intelligent reflecting surface, and makes a prospect summary. At the end of the paper, the open issues are identified as our future research directions.

A 3.3-kV silicon power diode with backside double injection holes is presented in this paper, which has been studied according to TCAD simulations. By setting the backside p-regions on the cathode side of the active area and setting the inside n+ layer in the backside p-regions, that is, the combination of P+N-P+N+ structure, P+N-P+ structure and P+N-N+ structure is set from the anode to the cathode. This new diode structure combines the advantages of the RFC diode and the CIBH diode, which could greatly improve the over-current turn-off ruggedness and soft reverse recovery characteristics at low current densities.

In order to solve linear constraint minimize dispersion problem in beamforming with non-Gaussian signals or noise, this paper proposed a real time adaptive beamforming technique based on modified conjugate gradient (MCG) method. Compared to conventional adaptive method, MCG have faster convergence rate than stochastic gradient (SG) method, and comparable convergence rate with recursive least square (RLS) method with lower complexity. And it simplifies the procedure of conjugate gradient method with one iteration per snapshots by using inexact line search scheme. Simulation has demonstrated the superior performance in output SINR and its convergence rate.

In this paper, we compared various deep learning methods in application of high-resolution remote sensing image classification. The accuracy rate of similarity in top 5 was more than 85% based on our database with the algorithms of VGG, ResNet, Inception, InceptionResNet, Xception, DenseNet and MobileNet. Among the seven models tested, Xception model achieves the highest model accuracy, the smallest training shock and the least training time.

Quantum communication is a theory based on the idea that quantum states can be transferred from one place to another. This theory has several advanced features. Specifically, it has high security, low time cost and low power overhead. Quantum information can make the execution of tasks much less efficient. However, much research has been done to achieve quality control by proposing utterly new security algorithms. The modulation and demodulation tasks of QC have not been sufficiently studied. Therefore, the modulation and demodulation methods are systematically summarised and analysed according to the two types of "signal type" and "modulation method type", respectively. In summary, PSK modulation outperforms OOK modulation in SRM-based multi-user detection schemes. The serious shortcomings of generalised secrecy modulation in terms of information-theoretic security mean that the security of GCM is not good. Moreover, there are many practical problems still required to be addressed. For example, the computational complexity characterisation and optimisation with correlation type receivers. Especially for ultra-high SE improvement have a large N parameter. Switching the slits at a reliable and efficient pace is a challenge. The high SE reduce the importance of speed. The complexity of the ML decoder combing with the joint modulation/coding should be explicitly modelled. Finally, the design of optical set-up in combination with amplifier structures recovering the loss due to MPD design.

In this paper, a multilayer tri-band microstrip antenna for the packaging system is designed based on half-mode technology. The antenna can work at 2.4 GHz, 3.5 GHz and 5.8 GHz at the same time, which can be applied to Wi-Fi and WiMAX frequency bands. The antenna uses three layers of rectangular metal patches stacked vertically as the radiation unit to ensure that the antenna will not interfere with each other in three frequency bands. The simulation results of the relative bandwidth of the three frequency bands are 0.083% (2.39 ~ 2.41 GHz), 2.29% (3.44 ~ 3.52 GHz) and 4.14% (5.69 ~ 5.93 GHz) respectively. The gain of the antenna is 1.77 dbi at 2.4 GHz, 3.49 dbi at 3.5 GHz and 5.4 dbi at 5.8 GHz which have good radiation characteristics. This antenna is advantageous because the structure is small and it has the independent tunable ability compared to other tri-band antennas.

Unmanned aerial vehicle (UAV) transmits data through wireless communication. Due to its small size and strong mobility, UAV can be used as a base station in the air to provide communication services for ground users. Therefore, UAVs can be used in natural disaster rescue, video shooting and other tasks. They can also carry some loads, such as high-definition cameras, infrared cameras, sensors, etc. to transmit and forward data through wireless channels. UAV technology has attracted more and more attention and has become one of the research hotspots in recent years. The cellular network of UAV is studied. The UAV channel is mainly based on line of sight link, and the characteristics of UAV channel transmission are analyzed. The influence of UAV position on the overall performance of the whole communication system is explored, and the transmission and control methods of UAV signal power are discussed.

In this paper, we study the resource optimization for multi-user orthogonal frequency division multiple access (OFDMA)-based broadband power line communication (PLC) system. In order to reduce the signalling overhead and enhance the network scalability, we focus on designing the distributed resource optimization strategy. Firstly, we formulate the PLC resource optimization as a network utility maximization problem. For solving the formulated problem, a two-step decomposition method is developed to decompose the formulated problem into a power control subproblem and a network configuration (subchannel allocation) subproblem. Then, we develop a continuity relaxation method for solving the optimal solution of the power control subproblem. In order to overcome the difficulty of combinatorial nature of subchannel allocation subproblem, we propose a Markov approximation framework to obtain the near-optimal solution. Finally, a distributed joint subchannel allocation and power control algorithm is proposed for optimizing the PLC system energy efficiency. Simulation results validate the theoretical analysis of our proposed scheme.

With the development of metro, the information safety of train control system has become a key element in closed transmission. This paper proposes an improved railway signal safety protocol I (RSSP-I) based on commercial cryptography 4 (SM4) for metro train control system. RSSP-I is mainly used for safety-related communication in closed transmission system and cannot defense the disguise threat. This paper improves RSSP-I protocol based on SM4 to defense the disguise threat, and can protect against repeat, loss, insertion, disorder, corruption, delay and disguise threat. The protocol proposed in this paper can strengthen the security of information transmission process.