This PDF file contains the front matter associated with SPIE Proceedings Volume 12918, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Due to the extremely high symbol rate and large bandwidth, orthogonal frequency division multiplexing (OFDM) signal on millimeter wave (mmWave) band often suffers a heavy computation and hardware resource burden for timing and frequency synchronization. Meanwhile, they also bring in a long processing latency, and lots of memory resources have to be paid to receiver. In order to deal with the above complexity and resource limitation problems of mmWave OFDM, many efforts have been taken in transmitter and receiver respectively: in transmitter, a training sequence (TS) with repeated properties is well designed, and it could be able to provide coarse and fine time delays and frequency offsets easily; in receiver, by using the signal structure, the receiving signal can be divided into several parallel transmission, and a two-stage timing and frequency synchronization method is proposed and realized on our field programmable gate array (FPGA) platform. Both the simulation and FPGA implementation show that the proposed method has a good synchronization performance with very low latency and hardware complexity.

Based on the service-oriented nature that 5G possesses, network slicing can build logical virtual networks in which various services are supplied by a common physical infrastructure. The function of network slicing is decided by the management of multidimensional resources that multi-access edge computing (MEC) offers. This paper presents the optimization of communication and computing resources in blockchain-enabled MEC network slicing. Given its immutability and traceability, blockchain is used to keep records of information concerning resource optimization, including resource information, system utility and slice strategy. The optimization goal of the resource allocation problem is to ensure the quality of service while maximizing the utility gained by the mobile network operator. The optimal resource allocation strategy is obtained using the deep deterministic policy gradient algorithm. The availability of the algorithm is proved through numerical simulations.

The diagnosability of many interconnection networks under the well-known PMC model and MM* has been examined as a gauge of the multiprocessor systems’ diagnosis capacity. The symmetric PMC model (SPMC model), a variation of the famous PMC model, is recently proposed by Zhu et al. Based on the PMC model, the SPMC model is obtained by adding two assumptions: (1) Each pair of neighboring nodes is believed to test each other; (2) For a given syndrome, only its minimal consistent faulty sets are to be located in 1-step diagnosis. As well-known interconnection networks, a number of characteristics of the 3-ary n-cubes have been examined. In this paper, the diagnosability of 3-ary n-cubes under the symmetric PMC model is determined. The result shows that the diagnosability of a 3-ary n-cube under the SPMC model is almost double that of the PMC model.

A sharpness evaluation method based on image edge detection is proposed for the problems of poor real-time, weak antiinterference ability and low sensitivity of the traditional gradient evaluation algorithm in the auto-focusing process of digital image processing. The edge pixel points are extracted from the image using an improved adaptive double-threshold Canny operator based on the idea of variance minimization within the OSTU class, and then the extracted image edge pixel points are evaluated. The evaluation function is based on the characteristics of human visual system, and the adaptability and sensitivity of the evaluation function to multiple direction edges are enhanced by improving the evaluation weights of the SMD function for horizontal and vertical direction edges and combining the advantages of the Roberts function for the evaluation of ±45° direction edges. The experimental results show that the improved edge gradient image sharpness evaluation algorithm proposed in this paper has the advantages of good real-time performance, strong anti-noise capability and high sensitivity.

The hydrogen market trading mechanism has problems such as high transaction costs and a small market scale, making it difficult to meet the demand for large-scale renewable energy hydrogen production. Blockchain smart contracts have the potential to solve these problems with their decentralized, distributed ledger, and transparent features. Firstly, this paper analyzes the interest relationship between trading subjects and introduces game theory to establish a game model and mechanism for hydrogen energy trading strategies. An improved ant colony algorithm is used to obtain the Nash equilibrium solution, providing the most optimal pricing strategy for trading parties. Secondly, by dynamically changing the stability coefficient of nodes and selecting new main node mechanisms, the probability of selecting malicious nodes is reduced, thus improving transaction consensus speed. Finally, through case analysis verification it was found that this proposed hydrogen energy trading mechanism can effectively reduce transaction costs and improve system operating efficiency. This research has certain significance in promoting the application of hydrogen energy in distributed energy.

Because of the great potential in deep sea exploitation, research of autonomous underwater vehicles (AUVs) has become a hot field of concern in academia and industry. As AUVs control technology has made significant progress in recent years, , the solution of path planning and path tracking has always been one of the research hotspots. The realization of path planning and tracking control in the AUV control center is crucial to ensuring its ability to perform complex underwater operations and maintain stable and reliable performance. This paper summarizes the main research progress of AUVs in the research area of path planning and tracking control and provides insights into the future development directions of path planning and tracking control.

With China’s booming economy, agricultural modernization has become an unstoppable trend. In order to solve the problem of real-time detection of multiple kinds of fruits, a real-time detection system of multiple kinds of fruits based on Raspberry Pi was developed. We trained the YOLOv5 network on a homemade multi-species fruit dataset. In the end, it not only realized the recognition of various types of fruits but also was able to output their location coordinates in the image. At the same time, we used the edge computing device Intel Neural Compute Stick 2 to realize the feed-forward inference operation on the neural network, so that the problem of low frame rate, when the Raspberry Pi performs realtime detection on a variety of fruits, is solved. In the real scenario, the real-time fruit detection model based on YOLOv5s network can effectively achieve the target detection of multiple types of fruits with 99.86% detection mAP, 99.93% testing accuracy and 10.37 FPS average real-time fruit detection frame rate. The system has many advantages, such as high accuracy, good robustness, fast calculation speed, etc. In addition, the embedded system can be further extended to realize the development of many intelligent products related to fruit recognition, such as intelligent fruit-sacking robots. Therefore, it can be concluded that the real-time detection system of multi-species fruits based on Raspberry Pi is helpful to improve the efficiency of farming, and the product can contribute to the modernization of agriculture in China.

With the rapid development of computer networks, the scale of networks has become extremely large and complex, and a slight fluctuation in network performance parameters may have an impact on the entire network service, which makes accurate prediction of network performance particularly important. In this paper, an improved grey prediction model is applied to the prediction of network performance, and a Markov residual correction model is used to correct the prediction results to improve the prediction accuracy. The experimental results show that the error between the predicted and actual values after Markov residual correction is significantly reduced compared with the traditional grey prediction model, and the results are closer to the actual situation.

Carbon emissions produced by logistics distribution have an obvious impact on the environment. A critical problem in logistics distribution research is the green vehicle routing problem. This paper investigates a green vehicle routing problem with pick-up and delivery and soft time windows for a single depot. The problem is developed as a single-objective mixed integer nonlinear programming model. The model aims to minimize fuel costs based on vehicle load and penalty costs for violating time windows while satisfying customer pick-up and delivery demands. This paper proposes an improved firefly algorithm based on chaotic initialization and differential evolution to solve the model. The results show that the proposed model minimizes fuel costs while minimizing the increase in penalty costs. Compared with a traditional heuristic algorithm, the improved firefly algorithm has higher solved precision.

To ensure security during the data encryption process, it is necessary for the RSA algorithm to find two large prime numbers before encrypting data and generate public and private keys. Since the traditional method is difficult to guarantee security when encrypting smaller data, we proposed a method based on Miller-Rabin prime number test to generate five prime numbers, and encrypt data based on the RSA algorithm of five prime numbers. The algorithm is implemented by using Python language, and experiments have shown that the improved RSA algorithm can significantly improve its security, with low time complexity. The algorithm execution is efficient, and the possibility of being violently cracked is also greatly reduced.

To address the current issues of path planning and limited applicability in patrol robots, an enhancement has been made to the existing A* algorithm, coupled with the utilization of the dynamic window approach, to design a global dynamic path planning method suitable for patrol robots. This approach involves introducing dynamic weights to enhance search efficiency, identifying collinearity among global path nodes to eliminate redundant points, and conducting obstacle detection between essential nodes to enhance path smoothness. An autonomous navigation algorithm is proposed, wherein the path generated by the A* algorithm is adopted as the global path. The essential nodes within this path are treated as local target points. By adjusting the initial pose angle and optimizing the weight coefficients of the evaluation function, the autonomous navigation algorithm’s optimization efficiency and obstacle avoidance capability against random obstacles are improved. Comparative simulation experiments are conducted to validate its robust performance in terms of planning efficiency, path length, and random obstacle avoidance. By establishing a ROS robot operating platform and conducting experiments in both static and dynamic environments, the autonomous navigation algorithm is shown to successfully plan superior paths and achieve real-time obstacle avoidance to reach the target point.

In recent years, the deteriorating atmospheric quality in major Chinese cities has posed significant threats to residents’ health. Effective analysis and assessment are crucial for addressing this issue. However, due to the lack of sufficient labeled data, conventional supervised learning methods often face challenges in application. Therefore, this study aims to analyze the atmospheric quality in major Chinese cities using a semi-supervised clustering approach. We collected five years of air quality data from major Chinese cities, including AQI, PM_2.5, PM₁₀, SO₂, NO₂, CO, and O₃. We applied the semi-supervised clustering method to cluster the data. Finally, we evaluated the model using the decision tree algorithm and calculated the accuracy of the model to be 89.28%. The findings indicate that the semi-supervised clustering approach is an effective method to gain deeper insights into the variations in air quality in major Chinese cities and to provide data support for formulating relevant policies.

Public film review is an important reference standard for film quality evaluation. 67.9% of users in China choose to use Weibo to share their personal film-watching feelings, and more than 80% of users say that word of mouth about a film on Weibo is easy to influence their film-watching behavior. This paper collected 236,011 Weibo topic data of 41 Chinese films in 2022. Sentiment time curve clustering model is used to explore the influence of sentiment changes on consumers’ film preferences and thus on film success. By combining the competitive distance measurement method and the partitioning algorithm, this paper selects the K-shape+SBD method with the best profile coefficient performance to cluster the sentiment time series data, and then the study found that the Weibo topic text clustering of 41 Chinese films was divided into four categories, showing four trends of “M-type”, “slowly descending”, “forward-inverted V-type” and “slowly rising” respectively. The changing trend of sentiment will significantly affect the success of films. Films with “slowly rising” are more likely to achieve commercial and word-of-mouth success, “M-type” and “forward-inverted V-type” are more likely to achieve better word-of-mouth, and “forward-inverted V-type” can attract more users to post film reviews.

In response to the increasing severity of hazy weather and the difficulty of prediction, an SSA-SVR based PM2.5 content prediction method is proposed. Specifically, the excellent search capability of the Sparrow Search Algorithm (SSA) was utilized to search for the optimal parameter combinations for the Support Vector Regression (SVR) machine. Firstly, the meteorological factors are dimensionalized using factor analysis. Then the prediction effect of PM_2.5 in Beijing is experimentally compared with the regression model constructed by other algorithms. The results show that SSA has stable global search performance and can effectively reduce the influence of SVR parameter selection on the generalization ability and regression accuracy of the system. This is useful for monitoring and prevention of haze and so on.

Aiming at the situation that the root-mean-square error exceeds the requirement of warfare technology index during the static angular accuracy test of high-precision explosive point measurement system, the target characteristics and imaging quality of the target light are analyzed in this paper. It is found that the binarized center-of-mass and form-center detection algorithms are inadequate. On the contrary, the sub-pixel subdivision method which is used to detect the center coordinates of the target lamp, has a good effect on the target lamp center location, solving the problem of angular accuracy.

Retransmission is a simple and effective method to improve transmission reliability in content delivery. However, wireless multi-hop networks have the problem of path instability. If the retransmission still uses the previous transmission path, it is highly likely that the retransmission will still fail. On the other hand, frequent transmission failures can lead to the need to frequently initiate routing protocols to rebuild transmission paths, thereby increasing network overhead and reducing network performance. NDN (Named Data Networking) is a content-oriented network architecture where nodes can request the same content through multiple FIB interfaces. Based on the content-oriented nature of NDN networks, this paper proposes a Dynamic Path based Retransmission (DPR) mechanism in wireless multihop networks. If a node receives a retransmitted Interest, it dynamically selects a FIB interface for forwarding based on the historical request failure rate and success rate from multiple FIB interfaces that can request the content. The experiment shows that the method proposed in this paper can effectively improve the success rate of content delivery.

This paper proposes an incremental object detection model framework based on knowledge distillation. The framework addresses the issue of catastrophic forgetting by optimizing the distillation loss function and expanding the labels of the new dataset based on the old object classes, which significantly enhances the recognition capability of the model for the old objects. The framework is implemented and evaluated on the two-stage object detection algorithm. Experimental results on the PASCAL VOC dataset demonstrate that the proposed model achieves competitive performance compared to state-of-the-art methods.

The optimization of reconnaissance constellation is an important step in the construction of space-based information system, and the coverage performance is the key index of the optimization of reconnaissance constellation. However, the coverage performance of the current constellation is mostly obtained through the simulation system, which has high time complexity and low efficiency, seriously slowing down the speed of constellation optimization. In order to improve the efficiency of constellation optimization, a two-stage optimization scheme of reconnaissance constellation based on machine learning is proposed. First, the trained machine learning prediction model is combined with the evolutionary algorithm to carry out rough optimization of the reconnaissance constellation, and the decision variable space after the scope is reduced is obtained. Then, in the reduced decision variable space, the simulation system and evolutionary algorithm are used to further optimize the problem, and the final result is obtained. The scheme proposed in this paper speeds up the convergence of population by reducing the search space of decision variables, thus improving the optimization efficiency. By comparing the optimization results of this scheme with those of the traditional scheme, it is found that the proposed scheme is 78.21% shorter than that of the traditional scheme under the premise that the quality gap between the optimization results of the traditional scheme and the traditional scheme is acceptable, which greatly improves the optimization speed of the reconnaissance constellation, and verifies the effectiveness and advancement of the proposed scheme.

The civil aviation industry is the fundamental industry of the national economy, and the good operation of airports, the reduction and resolution of complaints are important aspects of the development of civil aviation transportation. Based on basic data such as the number of civil aviation airport complaints in the past 10 years, this article proposes a modeling and prediction method for airport complaint volume based on ARMA model, and uses the AIC (Akaike Information Criterion, AIC) to determine the optimal order of the model. The prediction results show that the ARMA Model can still achieve high-precision prediction even when the order is not high, and the prediction results of (8, 1) and other 9 orders and above are consistent with actual data. The prediction results reflect the seasonal changes in civil aviation airport complaints, reflect the seasonal characteristics of civil aviation transportation, and have a high degree of confidence. Compared with the Least Square (LS) method and Gray Model (GM) prediction, the prediction method based on ARMA Model is more reasonable and has prominent advantages.

Mammography is main tests used for breast cancer risk assessment. However, due to the high similarity of the mass to breast tissue, blurring of the mass edges causes the task of segmenting the breast mass on mammograms to be challenging. To reduce computational costs and the workloads of radiologists, deep learning techniques based on computer vision have become a common implementation in the field of medical image segmentation. However, owing to the locality of the convolutional operation, the neural networks cannot effectively learn global and remote semantic information. We propose a novel gated axial transformer network (GATNet) framework for the mass segmentation of mammograms. GATNet uses an encoder-decoder structure. First, we use axial attention to decompose 2D self-attention into two one-dimensional self attentions. Second, we construct an efficient location-sensitive gated transformer module for image features to establish remote contextual dependencies. We conducted evaluations the proposed GATNet on two publicly available breast mass segmentation datasets. The average Dice similarity coefficients between the GATNet results and INBreast and CBISDDSM data were 80.98% and 83.63%, respectively. The experimental results indicate that GATNet effectively reduces both the computational and spatial complexity of the medical image segmentation network, demonstrating its remarkable performance.

Digital watermarking, as a technology to protect copyright, integrity, copy prevention or direction tracking of digital products, is currently commonly used to protect confidential documents and files within enterprises. In view of the PDF document format which is widely used in enterprise documents, this paper presents a method of PDF document watermark recognition based on natural language processing technology. By collecting a large number of PDF documents, using the improved N-gram language model based on forward and reverse matching algorithms to segment text content, a KenLM language model based on language model probability and conditional probability calculation rules is established to identify PDF document watermarks, which effectively improves the accuracy of PDF document watermark recognition. The validity of this method is verified by selecting a PDF format document of an enterprise, training the language model and calculating the prediction accuracy.

In the paper, high-resolution Synthetic Aperture Radar (SAR) imaging system is studied and its indicators are analyzed. To achieve high resolution, the system has a large transmission bandwidth, long synthetic aperture time and large azimuth scanning angle. According to the characteristics of the system, multi band fusion technology, large azimuth angle technology, high resolution imaging processing technology, and azimuth variable frequency technology are introduced. Also, the solution to the key technique of imaging is given in the paper.

Picture division has a special meaning for computer visualization and schema identification. Fast target extraction from deterministic images is an important problem facing real-time picture manipulation. Traditional areal models rely on globally converged messages to achieve fault-minimized segmentation. Its image segmentation is ineffective and takes up a lot of time. To address this problem, this paper proposes a model that Fast Region Image Segmentation of the Least Squares (FRISLS). Specifically, the target as well as the backdrop of the primary picture is approximated by just a pair of constants in order to establish the minimum error function. The weight matrix is used to reduce the influence of the background on image segmentation, and least squares are introduced to achieve fast convergence of the model. Through comparison with other area model-based approaches, it is validated the effectiveness of the study. The results indicate that this method ensures high precision of picture division, and enhances the performance of picture splitting efficiency.

Currently, rail transit has developed more and more intelligently, and intelligent vehicles are of significance in reducing traffic accidents and improving transportation capacity and have broad market prospects. This research combines infrared image acquisition and machine algorithms for obstacle recognition. The SURF algorithm performs feature extraction and description, and elimination of mismatched point pairs on infrared and visible light images, which can do image recognition on track obstacles. On this basis, a foreign object monitoring system consisting of an optical imaging subsystem, a lighting and detection light source subsystem, a pan tilt synchronous control subsystem, an image acquisition and processing subsystem, and an installation fixture and protection subsystem was designed. Finally, a demonstration application of foreign object monitoring under different working conditions in the section adjacent to the elevated line and highway of Guangzhou Metro Line 4, the open line section (K46+175) from Tung Chung Station to Huangge Auto Town station, for the 100 m section of the uplink and downlink double lines was carried out on the proposed system, which verified that the foreign object detection rate was 100% based on the test data, and the maximum detection response time was 4s (less than 5s). The relative deviations between the detected foreign objects and actual positions were less than 2.13% and the relative deviations between the detected foreign objects sizes and the actual sizes were less than 8.8%, which fully proved the effectiveness of the monitoring system.

Positioning of mobile robots is the most basic link of robot navigation, and also one of the key technologies for robots to achieve various complex tasks. This paper mainly studies the basic Monte Carlo positioning algorithm and adaptive Monte Carlo positioning algorithm, and analyzes the convergence process of positioning algorithm through simulation experiments. It is proved that the adaptive Monte Carlo localization algorithm has the ability to find the real pose of the robot in the map.

During operation, the shot blasting machine will generate a great deal of vibration and noise, which has a negative effect on both the staff’s working environment and the shot blasting machine’s ability to function normally. Therefore, this paper first performs a time-frequency analysis of the vibration signal obtained from the shot blasting shell and makes a preliminary determination that the shot blasting shell will be affected by various vibration sources. Then, based on the blind source separation technology, the time domain signal collected from the vibration sensor is transformed into a sparse signal by the Fast Fourier Transform (FFT). The analysis reveals that the motor vibration transmission and rotor imbalance are the shot blasting machine’s primary sources of vibration, with the impeller being the primary cause of rotor imbalance. Thus, noise reduction may be achieved by optimizing the impeller. The stability of the impeller will be increased, the noise of the shot blasting machine will be reduced, and the impact of the vibration of the shot blasting machine on the working and working environment of the shot blasting machine will effectively be reduced by adding alloy cobalt to the blades and some of the device groups to improve the comprehensive performance of the blades.

Compared with traditional encryption technology, physical layer security technology uses channel noise to realize the key with the same length as plaintext. Even if the eavesdropper has strong cracking computing power, it cannot decipher the confidential data without the key, which can truly realize “One-Time-Pad” and ensure the security of wireless communication. In this paper, we propose a physical layer security verification system based on software defined radio. In the system, base stations and legitimate users can obtain channel information efficiently, and eavesdroppers can use the powerful storage and computing resources of the cloud server to decrypt confidential signals without time constraints. Based on the system, this paper designs and verifies the physical layer security scheme with joint precoding and Artificial Noise, which fills the gap of the prototype verification platform of information-theoretic security and provides strong support for verifying the unconditional security features of the physical layer security technology.

In this paper, an ultra-wideband Ka-band low-noise amplifier (LNA) is presented using broadband transformer matching and gate-drain feedback technology. The proposed LNA achieves ultra-wideband while maintaining the high integration and low power consumption of traditional source-degenerated LNAs. Additionally, cascading technology is used to increase the peak gain. The proposed LNA is implemented in a 55 nm CMOS process. The simulation results show that the LNA achieves a stable gain response in the frequency band of 21.7-36.5 GHz, and the peak power gain is 20.8 dB. The minimum noise figure is 4.4 dB. The DC power consumption is 15 mW with a compact chip area of 0.33 mm².

Artificial noise, as a physical layer security technique, differs from traditional communication security mechanisms based on cryptography as it cannot be easily cracked by rapidly developing computational power. The theoretical research on the security of artificial-noise-assisted MIMO (Multiple-Input Multiple-Output) communication systems is quite comprehensive. However, the practical effectiveness of artificial noise in real-world systems still needs to be further validated through experimentation. This paper proposes a real-time platform for simulating experiments on MIMO communication systems assisted by artificial noise. By analyzing experimental data, it demonstrates that artificial noise has a slight impact on the reliability of legitimate MIMO communication systems while effectively enhancing system security.

This paper investigates the physical layer security in wireless communication networks, focusing on the scenario where a base station sends confidential messages to a legitimate node while facing potential eavesdropping. The study assumes that the channel state information for both authorized and unauthorized nodes is not available. To assess both security and reliability, the paper introduces the use of bit error rate (BER) as a quantifiable security metric. The objective of the research is to evaluate the security performance improvement of the system through BER. Specifically, the impact of artificial noise (AN) power allocation ratio, and the number of transceiver antennas on BER performance is characterized. The numerical results demonstrate that artificial noise is effective against unknown passive eavesdropping attacks and leads to significant performance gains. For instance, by transmitting artificial noise power at approximately 50% of the total power to deliberately impair the eavesdropper’s reception, the BER gap between intended and unauthorized users can reach 0.4.

Wi-Fi is a widely used wireless local area network (WLAN) technology, and its security issues have always been a matter of concern. The counter with CBC-MAC protocol (CCMP) is a commonly employed security protocol for Wi-Fi. However, vulnerabilities in the 4-way handshake process can allow attackers to carry out man-in-the-middle (MitM) attacks. In a successful attack, the targeted station (STA) connects to a rogue access point (AP) established by the MitM, which can have severe consequences for Wi-Fi users. This paper focuses on MitM attacks in the context of WPA2, CCMP, and WLAN security. We analyze and implement a MitM attack scenario, and based on this scenario, we summarize the potential hazards resulting from a successful attack. Furthermore, we present solutions to help prevent falling victim to such attacks.

This paper investigates the performance of adaptive modulation in unmanned aerial vehicles (UAVs) relay networks, in order to enhance the utilization of the spectrum resources and ensure the stability of the signal transmission, the UAV relay with the best communication link is selected to transfer the information to the target node via the decoding and forwarding protocol. on this basis, the outage probability of multiple unmanned aerial vehicles (UAVs) assisted communication is derived under independent Rayleigh fading channels. The simulation results indicate the availability of improving the performance of communication relays and communication networks and further suggest that adaptive modulation can significantly improve system performance.

In the context of modern information warfare, wireless communication technology has been widely used. The transmission mode of wireless communication is to use electromagnetic waves in the air to carry out information exchange. Compared with the traditional way of transmitting information through cables, wireless communication has better adaptability, expansibility, and easy maintenance. With the rapid development of wireless communication technology, the number of various electromagnetic equipment is increasing, and the communication signal system and modulation modes are becoming more and more complex and diversified. The overlapping of various electromagnetic signals makes the electromagnetic environment increasingly complicated. It has become a key problem that needs to be solved in the field of communication countermeasures to quickly discover signals from the broadband spectrum and locate important enemy signals through individual recognition, protocol analysis, and other technologies. In this paper, we propose a simple baseline approach for signal detection and extraction. First, we store the received IQ signal in a buffer pool. In order to realize real-time processing for broadband signals, we first perform blind estimation of signal to noise ratio (SNR), and filter out the signals based on the estimated SNR, and then only process the signals whose SNR is greater than a certain threshold. Then we detect the filtered signal by fast Fourier transform and threshold comparison from the frequency domain, and locate the time domain and frequency domain position of the signal. Our experiments demonstrate that our proposed approach can effectively detect the signals in real-time. When the SNR is greater than 3db, the signal detection error is not more than 0.01.

Microservice architecture is a new architectural pattern for deploying applications and services in the cloud, which aims to split a large service into a series of different and independent microservices, and provide users with better software design services. However, with the increasing business demand and the continuous expansion of the scale of the microservice system, the complexity of the system itself and the complexity of the network relationships make fault detection difficult. From the perspective of microservice node status and dependencies, we propose a novel fault detection algorithm based on a spatial-temporal convolutional network. It uses a graph convolutional network to effectively capture the complex spatial dependencies of microservice network relationships and constructs a temporal convolutional attention module, which uses dilated causal convolution, residual block and attention mechanism to better control the memory length of the model, obtain the key features in the time series, and improve the accuracy of the model. Finally, a large number of experiments prove the effectiveness of the algorithm in microservice fault detection.

Linear motor has extremely wide application in linear motion, but the end effect and cogging effect directly affect the thrust ripple of linear motor in working condition. In this paper, the flat permanent magnet synchronous linear motor (PMSLM) has been studied, and the cogging force has been modeled with parametric method in Maxwell, the influence of slot shape, pole-arc coefficient, and slot width to tooth width ratio (RST) on cogging force has been simulated, and the suitable structure parameter has been derived. Finally, the structure parameter of this motor is optimized, and the feasibility has been verified with simulation analysis.

In response to the threat of uncontrolled flying drones to the safe operation of railways, this article constructs a low speed and small UAV (unmanned aerial vehicle) control and management system based on radio detection and counter measures technologies. Based on navigation decoy, precise direction finding technology and, and link jamming technology etc., Effectively meeting the core requirements of rapid discovery, precise identification, and high efficiency disposal of small UAV along the railway, it has application value in scenarios such as safe railway operation and key location security.

In view of the development trend of manned spacecraft information system, especially the audio-video technology, combined with the requirements of manned spacecraft audio-video interaction, this paper proposes a system architecture and information processing flow of manned spacecraft audio-video processing system supporting CCSDS AOS protocol and the traditional digital information transmission protocol. Solutions are also given to some certain important issues such as system flexibility, robustness, and convenience. It lays a foundation for better support of space-to-earth communication for future manned spacecraft.

Because of the openness of the wireless relay channel, the confidential information transmitted in the channel is very easy to be eavesdropped on by malicious eavesdroppers and cause information leakage. In specifically, a friendly jammer will transmit a jamming signal to confound an eavesdropper attempting to listen in on a relay transmission. For wireless relay networks using the Rayleigh fading channel model, this research proposes a physical layer security technique based on cooperative interference. In the system, we design two-time stages to realize the transmission of the source to the user. During the initial phase, the signal source transmits a composite signal to both the user and the relay. In the second phase, the confidential information is relayed and artificial noise is generated by friendly jammers to confuse eavesdroppers. Through our proposed method, we derive precise analytical expressions for the intercept probability (IP) and outage probability (OP). The correctness of the theoretical derivation is verified by numerical results and theoretical simulation.

With the rapid development of 5G, networks are moving towards ubiquity, sharing, and intelligence. Traditional networks suffer from the problem of not being able to schedule routes accurately. The advent of intent-driven network (IDN) has made the network more flexible and efficient. IDN could identify and integrate UE intents in natural language and translate them into cloud network scheduling policy to manage cloud-network resources on the ONAP (Open Network Automation Platform), which is the platform for managing and automating network and edge computing services. To satisfy a lot of different demands, ONAP and IDN scheduling are necessary. MEC cloudifies the third-part App and saves the calculated resources in the local UE (user equipment). This paper introduces the containers and slices management, components of ONAP and IDN, introduces the MEC (mobile edge computing), and designs the procedure to manage the cloud-network user intent instances.

This system mainly utilizes the combination of VLC and WiFi communication technology to achieve a home wireless sensor network. The main design content is the information transmission of the link combining VLC and WiFi communication, as well as the transmission of sensor data in the network. Using this network as the medium, a more practical smart home system is achieved. The main control adopts STM32 Micro-Control Unit, which is the controller of the information transmission and sensor acquisition system. It completes the acquisition, processing, and analysis of the data involved in the visible light emitting module, visible light receiving module, temperature and humidity sensor, and then controls the humidifier and Whole-house fan according to the analysis results. Internet cloud platforms and mobile apps are responsible for receiving data, remote display, and remote control of system execution components. The client visualization interface uses the Internet of Things cloud platform, enabling users to easily know and control their home devices anytime and anywhere, greatly enriching and facilitating people’s home life.

In modern communication intensive networks, different transmission schemes have an important influence on the network transmission performance. Choosing a suitable transmission scheme can improve the efficiency and reliability of network transmission, and then improve the quality of network application and user experience. Therefore, it is of great significance to study and analyze the advantages and disadvantages of various transmission schemes and put forward corresponding optimization strategies. In this study, three transmission schemes, FTP upload, HTTP upload and UDP upload, are tested and compared in a large-scale communication intensive network. The test involves three aspects: upload speed, transmission time and data reliability. The authors tested in different network environments through the communication intensive network testing tool. The experimental results show that different transmission schemes have great differences in performance. UDP upload scheme has the best performance in upload speed, with an average upload speed of 23.81 Mbps, but the data reliability is poor, with an average data transmission reliability of only 88.3%. FTP upload scheme is the best in data reliability, with an average data transmission reliability of 99.2%, but the upload speed and transmission time are relatively slow, with an average upload speed of 8.57 Mbps and an average transmission time of 68.14 seconds. The HTTP upload scheme performs well in upload speed and transmission time, with an average upload speed of 12.79 Mbps and an average transmission time of 42.81 seconds, but the data reliability is only 98.6%. In practical application, the transmission efficiency and reliability can be further improved by adopting technical means such as bandwidth limitation and data compression.

The article investigates the flight scheduling problem during cooperative detection of UAVs and radars in a multi-station time-differential localization system, taking into account constraints such as the UAV’s cruise speed, full-fueled cruise time, loaded cruise time, and re-takeoff interval time. The time-series method is introduced, and the problem is decomposed into two sub-problems: the assignment of the task chains list and the assignment of UAV sorties, both of which are optimization problems belonging to the set partitioning. For the task chains list problem, the optimization model is established and an improved PSO algorithm is designed, including unconstrained treatment based on penalty terms, adaptive inertial weighting factors, and hybrid join mutation operation of GA algorithm. For the assignment of UAV sorties, the optimization model is established and a solution algorithm is designed. Simulations and analyses of the model and algorithm were carried out to prove that the model and algorithm are reasonable and effective.

This paper is based on MATLAB software and conducts mathematical modeling and simulation calculations of an engine equipped with a double rotary valve cylinder head assembly, which was developed by the project team. Structural parameters were selected based on the size of the prototype that had been tested, and the working process inside the cylinder was modeled and simulated. The simulation results were analyzed and compared with those of the original poppet valve engine. The results indicate that the intake endpoint pressure and the maximum burst pressure of the rotary valve engine are both improved compared to the original engine. This study provides a method for calculating the intake and exhaust processes of the rotary valve engine and can provide a reference for the matching of the rotary valve cylinder head assembly with various types of engines in the future. These results may contribute to improving the design and performance of the engine to meet the needs for higher efficiency and lower emissions.

Trajectory prediction is a crucial tool for analyzing vessel motion behavior, assessing vessel traffic risks, and planning collision avoidance routes for intelligent ships. We propose a novel deep-learning architecture-based model Trajformer to perform the task of vessel trajectory prediction. By preprocessing AIS data, utilizing the LSTM to learn local dependencies, and self-attention mechanism to enhance the capacity of dependencies learning, Trajformer achieves multi-step vessel trajectory prediction. Inspired by time series preprocessing and forecasting tasks, we also propose Trajformer+SD, a simple yet effective approach that applies Series Decomposition on the input sequence. Our modal can effectively model the vessel trajectory and make predictions, as demonstrated by our experiments on three real-world vessel trajectory datasets. Compared to the sequence to sequence (Seq2Seq) and sequence to sequence with attention (Seq2Seq_attn) model, our proposed model exhibits a better performance in prediction tasks.

This paper reviews the evolution of IT system architecture and highlights that container technology based on cloud computing is an inevitable trend in the future of IT technology. And big data and artificial intelligence are also closely related to container technology. For instance, the architecture for computing-storage separation and the distributed training of large models are uninterruptedly dependent upon container technology. Numerous container-based applications require robust management tool suites. Although Kubernetes is a well-known tool for container orchestration, its native scheduling is not tailored towards offline workloads, like big data and artificial intelligence. In this context, this paper first examines the principles of Kubernetes native scheduler and identifies the challenges encountered when dealing with batch tasks, then presents the concept, the workflow, and integration methods of batch schedulers, and finally summarizes its significant features and several open-source frameworks. Ultimately, enterprises’ IT infrastructure is continuously moving towards cloud and containerization. Comprehensive scheduling architectures that support both online and offline tasks are crucial, as they significantly impact the enterprise’s overall IT scalability, efficiency, and business outcomes. Therefore, this area has garnered the industry’s attention.

This paper investigates the problem of trajectory generation in a multi-station passive time-difference localization system in which an unmanned aerial vehicle (UAV) equipped with a passive payload replaces a secondary radar station in the localization task. Based on this, a method is given for segmenting the optimal deployment trajectory so that the deployment trajectory can be executed by the UAV according to the speed and load-on cruise time constraints of the UAV. Simulation and analysis of the model and method are carried out, the results meet the expected results, and these methods are reasonable and effective.

This paper presents a simulation of the in-cylinder flow field during the intake process of an engine equipped with a dual-rotating valve cylinder head assembly, which was designed and developed by the project team. Based on the dimensions of the original prototype, a three-dimensional numerical model was established. Using Starccm software, the simulation calculation of the intake process flow field of the engine was carried out, and the results showed that the intake state of the dual-rotating valve system was stable and the flow field had no severe disturbance, which was beneficial to the engine’s intake. This paper proposes a simulation calculation method for the rotating valve engine, which has a certain reference value for the research and development of the new type of valve train system and its adaptation to different engines.

This paper presents a method toevaluate the potential of data centers to involvement in power demand response, which is crucial for maintaining the equilibrium of supply and demand in the grid, promoting clean energy consumption, and mitigating energy crises. The data center is a valuable demand response resource due to its flexible adjustment methods, large load volume, and strong controllability, facilitated by its spatio-temporal scheduling mechanism. The proposed method considers multiple types of flexible resources and includes an analysis of data center participation demand response mechanisms, an evaluation model based on time series simulation optimization operation, and practical examples to verify its effectiveness.

In large-scale private computing applications, various arithmetic relations appear as or can be reduced to matrix relations. In this paper, we establish the efficient zero-knowledge proof (ZKP) for the quadratic matrix relation over finite field Fp with two witness matrices. In private computing tasks, lots of typical relations are instances or special cases of this form, e.g., matrix multiplicative relation, inverse relation, isometric relation, etc. Different from the widely-applied vectorspecific method, our method is matrix-specific. The matrix equation is treated as a tensor equality and probabilisticequivalent reduction techniques are applied to reduce the non-linear matrix relation to simple vector relation. To the authors’ best knowledge, currently, there are no matrix-specific methods to ZKP for nonlinear matrix relations. Compared against the current (vector-specific) method, our method substantially outperforms it in all critical aspects, e.g., for n-raw t-column matrix witnesses, the required size of common reference string (c.r.s.) can be compressed by a factor of 2nt and the number of messages, group and field elements are all reduced by a factor of ≈2 for large-size witnesses. Computational complexities in both methods are almost the same.

In large-scale private computing applications, various arithmetic relations appear as or can be reduced to matrix relations. In this paper, we establish the efficient zero-knowledge proof (ZKP) for the quadratic matrix relation over finite field Fp with three or four witness matrices. In private computing tasks, lots of arithmetic relations are instances or special cases of such form, particularly some matrix structural decomposition relations. Different from the widely applied vectorspecific method, our method is matrix-specific. The matrix equation is treated as a tensor equality and probabilisticequivalent reduction techniques are applied to reduce the non-linear matrix relation to simple vector relation. To the authors’ best knowledge, currently, there are no matrix-specific methods to ZKP for nonlinear matrix relations. Compared against the current general linearization (vector-specific) method, our method substantially outperforms it in all critical aspects, e.g., for n-by-t matrix witnesses the required size of common reference string (c.r.s.) can be compressed by a factor of 2nt and the number of rounds, group and field elements in messages are all decreased by a factor of ≈2 for large-size witnesses. Computational complexities are almost the same in both methods.

With the increasing application of composite materials in aircraft structures, bolt connections are widely used due to their high reliability, strong bearing capacity, and easy disassembly. However, the forming accuracy of composite components is lower, and inconsistent deformation and assembly stress is easy to occur, which affects the shape accuracy and safety of aircraft seriously. In this paper, taking the assembly of a composite plate and L-shaped stringer as an example, the effects of horizontal clearance between composite components, bolt tightening force, and pin-hole fit clearance on the assembly deformation were analyzed through finite element orthogonal experimental design. The results showed that the average deformation increased with the horizontal clearance, and the average deformation decreased with the tightening force. The finds also demonstrated that pin-hole fit clearance had little effect on the assembly deformation, and the gap filling method was an effective way to restrain the assembly deformation.

The numerical simulation technology is mainly through the computational fluid dynamics (CFD) method, which can quickly and accurately simulate the aerodynamic performance of the vehicle at a specific speed, such as air resistance Cd, downforce Cl, etc. These simulation results can provide important references for automotive design and optimization, reduce trial and error costs, and improve automotive design efficiency. In this paper, CATIA software is used to design the simplified three-dimensional model of the car. Based on the basic theory of fluid mechanics, AnsysWorkbench software is used to analyze the aerodynamic simulation of the car in driving process. The data on resistance and downforce are obtained and analyzed for reference by car design department.

The purpose of this paper was to evaluate the effect of four kinds of marginal finish line design on stress distribution in posterior zirconia all-porcelain crown prosthetic system. Four 3D finite element models with different margin designs of mandibular first molar restorated by zirconia all-porcelain crown were developed. Each model was designed with one kind of finish line margin respectively (SC0.5: shallow concave finish line with a width of 0.5 mm; SC1.0: shallow concave finish line with a width of 1.0 mm; RA0.5: right angle finish line with a width of 0.5 mm; RA1.0: right angle finish line with a width of 1.0 mm). A load of 670 N which equals the greatest masticatory pressure in centric occlusion was applied to the zirconia all-porcelain crown vertically. The other load of 220 N simulating the lateral occlusion was applied at 45 degrees to the long axis of the abutment tooth. Then FE analysis was performed in each situation to observe stress distributions and Von Mises stress values in each layer of the zirconia prosthetic system which consisted of zirconia crown, cement layer, and abutment tooth. For all models, stress was localized in the occlusal loading area and the boundary of the crown. The minimum stress values were obtained in model SC1.0, the greatest stress values occurred in model RA0.5. Furthermore, the Von Mises stress values of the SC group were apparently higher than that of the RA group. The FEA method can effectively evaluate the stress in finish line. The results suggested that a combination of shallow concave finish line with a width of 0.5 mm potentially have more suitable geometry to reduce stress concentration.

The clutter generated by multiple transmitting pulses overlaps in the time domain, which results in the two-dimensional aliasing of the range domain and Doppler spectrum of the clutter for the medium and high repetition frequency radar. In the simulation of different battlefield environments, a large number of range-Doppler grid units will be generated. It is necessary to establish a simulation model of aircraft based on GPU-CPU architecture to solve the track, attitude and radar parameters in real time to generate grid units, and then add antenna and sea situation information to calculate the cross-heavy sea clutter. This is of great significance for rapidly generating large span heavy sea clutter and improving the speed and reliability of sea clutter simulation.

With the widespread application of blockchain technology on the Internet of Vehicles (IoV), numerous Blockchain-based IoV (BIoV) solutions have emerged. As one of the important application scenarios of IoV, highway needs a highly reliable reputation management scheme to ensure secure data sharing. To solve the problems of privacy leakage and hostile attacks caused by the lack of trust between nodes in the highway BIoV scenario, a reputation management scheme for Highway Internet of Vehicles based on subjective logic (HISL) is proposed. Under the condition of low transmission rate of highway packets, real-time reputation update can effectively prevent malicious nodes. In addition, our system model ensures privacy security and data integrity in the highway IoV environment and effectively improves data sharing efficiency. The performance analysis results show that the HISL solution can detect the disguised malicious nodes in a shorter time, thus effectively solving the security problem in highway and even more IoV scenarios.

In order to realize the machine-independent decision-making and control of ship intelligent navigation, this paper explores and builds a universal and practical ship-driving machine language system, analyzes the command requirements of driving behavior and ship state according to the three collision avoidance behaviors of ship navigation, and creatively designs the coding form of ship driving behavior index and ship state index by using 01 code and ILP method, so as to provide a man-machine and machine-machine collaborative dialogue for ship intelligent navigation. And the intelligent development of shipping provides important basic theory and technical support.

Global maritime accidents have a wide range of spatiotemporal differentiation characteristics, and it is an interesting problem that can be tried to seek efficient and scientific algorithms to recognize the spatiotemporal differentiation characteristics of global maritime accidents. Through the analysis of the data structure of marine traffic accidents, a global spatiotemporal differentiation analysis method for marine accidents based on accident point density clustering is proposed. A two-dimensional kernel density algorithm and a density-based clustering method with noise are introduced, and a clustering fusion algorithm based on spatial meshing and point density is established to reveal the spatiotemporal differentiation characteristics of global maritime accidents. Combined marine traffic accidents with the spatial and temporal information obtained on IMO GISIS in the past 30 years, the cluster fusion algorithm is applied to reveal spatiotemporal differentiation characteristics. The results show that the data efficiency of global maritime traffic accidents has been greatly improved with the wide application of AIS. The global maritime accident hot-spots are basically stable in Western Europe, the pan-Mediterranean and East Asia, South Asia, etc. Greater cooperation among neighboring port states, can effectively improve maritime traffic safety, especially in maritime search and recuse cooperation. The clustering fusion algorithm based on spatial meshing and point density can be further applied to the scenarios application in the regional scope, so as to clearly reflect the temporal and spatial differentiation characteristics of maritime accidents in the scenarios.

To solve the problem of irregular and inefficient case characterization and discipline in the field of discipline inspection and supervision, a method of case characterization and discipline measurement based on knowledge map representation learning is proposed. The method uses the knowledge map of discipline inspection cases as auxiliary information, and the knowledge representation learning model is used to characterize the entities and relationships in the knowledge graph, and the results of the characterized representation are subsequently processed by similar case calculation, and then uses the results of the representation to calculate similar cases to further realize the qualitative discipline of cases. The negative sampling method of the knowledge representation learning model is investigated to improve the knowledge graph vectorization accuracy and facilitate the manipulation of the knowledge graph in the case of qualitative quantitative discipline task. The knowledge graph inference-based approach to case characterization and discipline improves accuracy and computational efficiency. By improving the existing model, the improved model can reduce the impact of noisy data to a certain extent.

This paper focuses on the field of emotional interaction, specifically using text data to evaluate expressed emotions. It aims to assess the emotional state of humans in order to determine how robots should adjust their behavior. To detect emotions in text data, various methods and algorithms can be utilized. In this study, the authors combine dictionary methods with machine learning algorithms for emotion detection. However, emotion labels are often vague and subjective, leading to a multi-label problem. To address this, the paper conducts experiments using naive Bayes, support vector machine, and neural network classifiers. Although the accuracy of emotion classification is low, the study demonstrates the importance of specific words in expressing emotions.

Due to the uniqueness of “one country, two systems” and “two customs territories” in Guangdong (GD) and Hong Kong (HK), cross-border road cargo clearance between the two regions requires the creation of several documents. These include the Road Cargo Manifest, Vehicle Declaration, Customs Declaration, HK Seamless, HK Bundle, and HK Customs Declaration, as well as additional documents such as Packing List, Contract, and more. The use of Artificial intelligence (AI) can streamline the process of generating these documents and ensure accuracy and efficiency in cross-border road cargo clearance operations. This paper focuses on four aspects: text scanning and recognition, extraction of commodity classification declaration elements, conversion of customs declaration documents, and intelligent commodity classification. These areas of research aim to address the time-consuming and labour-intensive process of document creation and verification. By leveraging AI technologies, such as optical character recognition (OCR), data extraction algorithms, and machine learning (ML), it is possible to automate and optimize the production and review of these documents, thereby reducing time and manpower costs in cross-border road cargo clearance operations between GD and HK. Based on big data analysis of customs declaration documents from both regions, AI technologies such as feature extraction and machine learning can be utilized to achieve automation in the identification of documents like road cargo manifests and customs declarations. This research applied to the China International Trade Single Window (Shenzhen) and significantly improved the efficiency and accuracy of customs declaration in GD and HK.

Segmenting tumors from brain MRI sequences is critical for prognosis assessment, radiomics analysis, surgical planning, and pathological diagnosis. However, semantic segmentation based on convolutional neural networks is on the regular grid representation of the input and output of the network, which leads to the tendency of the predicted label edges to smooth. The infiltrating nature of malignant gliomas also poses challenges for the segmentation task. Therefore, we propose an automated, standardized method for fine-segmenting brain tumor boundaries. This method introduces the PointRend module based on 3D-UNet and trains a model with coarse-to-fine segmentation on the open dataset BraTS. The performance of our model on the public validation dataset is as follows: Dice of 0.84699, 0.84640, and 0.84661 for the whole tumor, tumor core, and enhancing tumor area, respectively. The evaluation shows that our model produces a good and balanced performance for different tumor subregions.

Artificial intelligence is currently in an era of change, not only changing the artificial intelligence technology itself but also changing human society. It has become more and more common to use artificial intelligence as the core human-computer interaction technology to replace manpower. Intention recognition is an important part of the human-machine dialogue system, and deep learning technology is gradually being applied to the task of intent recognition. However, intent recognition based on deep learning often has problems such as low recognition accuracy and slow recognition speed. In response to these problems, this paper designs a BERT fine-tuning to improve the network structure based on the pre-training model and proposes new continuous pre-training goals. To improve the accuracy of intent recognition, a method based on multi-teacher model compression is proposed to compress the pre-training model, which reduces the time consumption of model inference.

The dissipative observer design for a singular neutral system is studied in the article. The rigorous dissipative analysis of a singular neutral system is carried out. And the observer-based dissipative control problem is discussed through the use of linear matrix inequality (LMI). The method of observer design is given. Finally, the proposed method's effectiveness and universality are demonstrated by a final example.

Through the study of users’ mental model, this paper discusses the interactive process of image retrieval, and provides ideas for accurate image retrieval in line with user psychology. Through the methods of concept list and user log, this paper builds a Users’ Mental Model in Image Retrieval (UMMIR), discusses the dimensions, characteristics and types of users of image retrieval, and applies UMMIR to the retrieval strategy, constructs the topological relationship of Tagcloud, and verifies it with the visual prototype. The UMMIR consists of three levels: the Platform Technology Level (PTL), the Interactive Experience Level (IEL), and the Image Content Level (ICL). It includes 12 dimensions: Service Platform (SPD), Intelligent Drive (IDD), Intellectual Property (IPD), Retrieval Entrance (RED), Feedback and Sifting (FSD), Interactive Interface (IID), Topic (TD), Emotion (ED), Composition (CD), Style (SD), Purpose (PD), Quality (QD). UMMIR has the characteristics of Scenarios, Complexity, Imperfection, and Gradualness. Users in UMMIR can be divided into five types: Target-driven, Art-driven, Structure-driven, Platform-dependent and AI-yearner. The visualization of UMMIR to the retrieval platform provides a direct basis for the optimization of the interactive process in image retrieval.

In recent years, with the continuous increase of renewable energy installation capacity, China has been facing increasingly severe problems of abandoned wind, abandoned light and abandoned water. Hydrogen energy as a feasible solution to this problem has attracted widespread attention. As an emerging secondary energy source, hydrogen faces trust and efficiency challenges in transactions due to its short development time and immature market, urgently requiring an optimized transaction mechanism and a trusted and efficient trading environment to adapt to complex trading markets. Therefore, this paper constructs a blockchain-based hydrogen energy trading system. Firstly, it proposes a mechanism for generating consumption vouchers for hydrogen energy, transactions and traceability. Then it designs the overall architecture system based on blockchain technology with core capabilities such as efficient consensus mechanisms, trustworthy transactions, identity authentication and credit evaluation to build the hydrogen energy trading system. Finally, the testing is conducted on multiple devices with multiple nodes deployed, which verifies the completeness and practicability of the system.

For the resilience issue of command information system, this paper explores the influencing factors of resilience and constructs an optimization model for operational activities allocation based on functional redundancy between system nodes. By optimizing the process of redistributing activities after system damage, the logical structure of the system is adjusted to improve the completion of combat tasks, thereby improve the resilience of the command information system.

Recent years, the related research on facial expression recognition is increasingly attracting people’s attention. As an important branch of artificial intelligence, facial expression recognition has a significant role in human-computer interaction and other related fields. However, due to the high similarity between different expressions and the small interclass differences, it is still a daunting task for researchers to obtain a robust and accurate expression recognition model. Considering the characteristics of human expressions, a new convolution mechanism is proposed, which we call the attention convolution module. We also apply the convolution module to the field of expression recognition and propose a new expression recognition network (ATB-NET). The experimental results show that our method can focus on relatively important regions in the face, besides, compared with previous work, it also has a significant improvement on multiple public data sets (such as FER2013, CK+).

With the development of computer technology and the urgent need of the commercial society, the recommended products urgently require the realization of intelligence and precision. The commodity intelligent recommendation system based on data mining uses Spring Boot technology at the front end and SQLite database at the back end. Data mining algorithm is used for data analysis and screening. Design and implementation the function modules of the commodity intelligent recommendation system based on data mining, such as home page, mall, exploration and me. Through the test, the performance and throughput rate of the commodity intelligent recommendation system based on data mining are good. It has good stability performance and achieves the goal of intelligent product recommendation.

Under the background of the “14th Five-Year Plan” to promote the digital level of public culture, this study discusses the knowledge organization and visualization of minority culture. The aim is to enrich the knowledge base of public culture and improve the digital level of public culture by constructing the knowledge graph of minority festivals. In this paper, the festival culture of the Shui nationality and Bouyei nationality is selected as the research object. Based on ontology theory, the relevant data is structured and processed, and then the data is visualized by the knowledge graph tool. This study successfully constructed a knowledge graph of Shui nationality and Bouyei nationality festival culture, which contains 353 entities and 951 relationships. The results show that the method proposed in this study can be effectively applied to the construction of a knowledge graph, and reveal tacit knowledge from it, providing support for the digitization of public culture.

This study introduces a hardware acceleration design scheme for real-time object classification using SDSoC development software, and compares the performance advantages and disadvantages of different optimization strategies. Unlike the traditional RTL design and HLS design, SDSoC adopts the whole system optimization compiler, which can provide automatic data connection and other functions, so that developers can invest more energy in algorithm optimization, reduce the development difficulty and speed up the development time. This study uses a convolutional neural network to classify objects and optimizes the network according to hardware performance. A real-time video acquisition and display system was built in Vivado. This study builds a complete system in SDSoC, and focuses on optimizing convolution operation and data transmission path according to the running time of each function in the builtin TCF profiler. The effects of different acceleration function settings and network quantization on performance and resource utilization were studied. Finally, different optimization strategies are compared and evaluated, and the best design scheme achieves 3.599 W power consumption and 9.24 GOP/s per watt performance.

The tropical cyclone (TC) is a strong and highly destructive tropical low-pressure vortex that often brings disasters such as strong wind, heavy rain, and storm surge. The formation and intensity forecast of TC is very important for TC disaster warnings. In this paper, we propose a three-dimensional temporal-spatial (3D-TS) attention TC forecast model based on deep learning, which considers the temporal-spatial relationship between TC variables on the basis of machine learning methods. The model introduces 2D convolutional neural networks (2DCNNs) and 3D convolutional neural networks (3DCNNs) to learn oceanographic variables and atmospheric variables of TC, while utilizing Long Short-Term Memory (LSTM) for capturing the temporal correlation in TC’s evolution process, and introduces 3D-TS to grasp the temporalspatial characteristic of TC and enhance the model’s precision. Through experiments, it was found that the model’s performance in TC formation and intensity forecast is better than many existing numerical forecasting methods, statistical forecasting methods, and machine learning methods. We validated the model on the Western Pacific TC dataset, in the TC formation forecast experiment, the model can achieve a maximum accuracy rate of 86.1% and an AUC value of 92.5%. In the TC intensity forecast experiment, the minimum error is 7.1 kt, showing that we achieved state-of-the-art results.

Lane detection, a fundamental component of autonomous driving, has recently gained significant traction in research and development. Recently, there are lots of methods introduced to improve the performance of detection by designing complex network. However, they ignore the potential application of the semantic parsing ability of large vision models in lane detection. In this paper, we propose a novel and simple detection approach, named Prior Lane Detection (PLD), to perform precisely detection. To be specific, our approach involves utilizing a large vision model to extract semantic proposals as prior information. The large vision model, trained on a vast number of images, exhibits robust zero-shot performance and possesses the ability to generate high-quality object masks. Then, we apply image and semantic proposals to train the detection network with the aim of improving its performance. Our PLD approach utilizes prior semantic information extracted from a large vision model to guide the network’s attention towards important regions during lane detection. The experiments show that leveraging the large vision model can enhance the performance of lane detection models. The results of the experiments prove the usefulness of the proposed PLD on the TuSimple dataset.

Cut tobacco width plays an essential role in the production of the cigarette industry. Current cut tobacco mainly has a width in the range of 0.9 mm, 0.95 mm, and 1.05 mm. With the increase of cut width, the proportion of medium, as well as long filaments in the final products, will increase. In contrast, when the proportion of short and broken shreds decreases, the structural distribution stability of the final products will increase as well. However, if the cut leaf is too wide, the rolled cigarettes tend to be empty, resulting in low flammability and slow ignition speed. Therefore, an online closed-loop control system of cutting machines based on high speed and accurate measurement of the cut tobacco width is a valuable but challenging problem in the modern cigarette industry. This paper proposes a purely nondestructive width measurement method by applying machine vision techniques and machine learning methods for industrial applications. First, the skeleton method and Hough line detection are integrated to achieve a sub-pixel measurement of cut tobacco. Then a global calibration method is implemented to map the measurement result from the image coordinate to a real-world distance. Our experimental results demonstrate that the proposed tobacco width measurement method has the characteristics of a low error rate as well as real-time execution efficiency.

The 14th Five-Year Plan proposes the objective of expediting digital development and constructing a digital China, aiming to minimize the reliance on physical travel while fostering increased data flow and sharing. However, the implementation of data privacy regulations, such as the Personal Information Protection Law, has resulted in conflicts between data sharing and privacy. To address this challenge, federated learning technology has emerged as an effective solution that enables data sharing while safeguarding privacy. Nonetheless, trust issues arise among federated members in relation to each other and central servers. This is where blockchain technology, with its notable features of verifiability, auditability, and traceability, can play a pivotal role in resolving trust concerns within federated learning. Consequently, a novel framework called trusted federated learning based on blockchain has emerged. This article initially defines the trust issue in federated learning and subsequently provides a chronological overview of research advancements in the domain of trusted federated learning based on blockchain. Furthermore, it discusses the advantages and disadvantages of existing research across three categories: blockchain structure design, consensus mechanism, and smart contracts. Additionally, the article highlights the application characteristics of this framework in fields such as the Internet of Things, connected vehicles, and medical services. Finally, the core challenges and future research directions of trusted federated learning based on blockchain are outlined.

Classifying the equipment type of hostile targets in military applications in SAR images is crucial and requires higher classification accuracy. Using data enhancement techniques, this paper conducts an experimental study on the classification of MSTAR dataset using a convolutional neural network with the addition of an attention mechanism and obtains the advantages of strong generalization ability, fast training speed, and high fitting efficiency of the target classification method proposed in this paper.

At present, most deep networks in text detection tasks have a prominent effect on regular text detection, but a poor effect on curved text detection. This is because most of the curved text has irregular lengths. When the text is curved, the length is often irregular, some paragraphs are elongated, some paragraphs are shortened, and the inclined Angle of the curved text is large, so the boundary box cannot fit the text area well in text detection. However, for Chinese seals, the Angle of text bending is difficult to be fitted with the commonly used text detection box, so this paper uses OpenCV to carry out polar coordinate transformation, and converts the original curved text into regular text so that the detection box is easier to fit for the next step of recognition. However, after polar coordinate transformation, some characters will be divided and thus cannot be recognized. This paper further proposes to use text detection to process the transformed boundary box, so that the characters can be accurately recognized without being divided after polar coordinate transformation. Through experiments, the recall rate of character recognition inside the official seal can reach 92.23%. However, only 27.32% of official seal characters without processing are recognized, which indicates that the recognition effect after transformation by polar coordinate method is good.

There has been rapid progress in the development of object detection methods. A challenging problem of deep neural networks based object detection algorithms is they rely heavily on abundant training samples. When dealing with scenarios where samples are scarce, the accuracy of these models can degrade sharply, thus hampering their final efficacy. Recently, the emergence of few-shot detection methods has addressed this issue. The majority of existing research in the literature concentrates on constructing few-shot detection models based on two-stage object detectors due to their high capacity and compatibility with affiliated structures. As a result, these methods are capable of delivering high-performance object detection with a limited number of training samples. However, two-stage backbones often imply a lower inference efficiency as well as applicability with deployment on edge devices. To realize high-speed object detection specifically in computation resource-constrained scenarios, we propose a one-stage few-shot object detector by integrating a meta-learning structure with YOLOv3 model. Experimental results demonstrate that our proposed few-shot object detector achieves comparable detection accuracy to existing two-stage object detectors under the same training conditions. Meanwhile, it significantly prompts the inference speed, enabling real-time object detection in videos.

In the context of intelligent vehicle lane changing, a trajectory with superior performance is typically planned in accordance with the constraints of the lane change path. This allows the intelligent vehicle to execute the lane change along the preplanned trajectory. Model Predictive Control (MPC) stands out among other control methods due to its ability to incorporate a variety of constraints into the control process. This ensures that the model aligns with the actual situation, rendering the control results practically significant. Consequently, this study employs the MPC method to design the lane change path tracking controller for intelligent vehicles. The designed trajectory tracking model predictive controller is validated in a simulation environment, demonstrating effective tracking control at various vehicle speeds. Furthermore, it can execute tracking control of the vehicle’s lane change path in accordance with the planned ideal reference trajectory.

Using deep learning to study the aging and damage classification of high-chromium martensite microscopic metallography requires a large amount of labeled data; usually a pre-trained deep learning model is used to deal with this problem, however, the pre-trained model contains a large number of available Features, although fine-tuning the entire pre-training model improves the model effect, it consumes a lot of computing power. In view of this, based on migration learning, this paper divides the VGG network into several parts according to the order of network reasoning from the perspective of domain adaptation, and divides and identifies the functions of each part from the perspective of down sampling; using high-chromium martensitic metallographic aging Each part of the network is fine-tuned separately with the impairment dataset. The research results show that the fine-tuning part of the model converges faster and saves more computing power than the fine-tuning of the entire model; different functional parts of the model have different effects on the final effect of the model, and fine-tuning the module closest to the input has less impact on the final effect of the model. In this way, a more efficient domain adaptation strategy for the aging and damage classification model of high-chromium martensitic heat-resistant steel is obtained.

Behavior recognition is an important research domain in computer vision. Extracting human motion features and identifying behaviors from video can provide important support for education, medical treatment, security, and other fields. However, current mainstream behavior recognition methods prove ineffective in the challenging context of recognizing students’ classroom behavior due to scene complexity, numerous interfering factors, and lack of public datasets. To address these limitations, this paper presents a novel recognition model that integrates behavioral pose information and attention mechanisms. The proposed model employs a residual convolutional neural network with an attention mechanism to extract spatial and temporal features from teaching videos while leveraging behavioral pose information to enhance the description of visual cues and capture high-level behavioral information embedded in depth features. To assess the model's effectiveness, we create a comprehensive dataset of students' classroom behavior and conduct extensive experiments. The results demonstrate the proposed model's high performance in recognizing college students' classroom behavior.

Considering the complexity of the driving environment and the high uncertainty of road conditions in autonomous driving, this paper proposes a real-time object detection model YOLOv5st, which is the integration of the advantages of YOLOv5s and Swin Transformer to achieve image recognition and object detection tasks in autonomous driving environment perception. The proposed model YOLOv5st is tested on the Transportation of Wuhan China (TOWC) traffic dataset, and the results show that the mean Average Precision (mAP) of the model is 89.8% in the validation set, which is a 2.0% improvement compared to YOLOv5s. And the average inference time of a single image of the new model is 20.6 ms, which can meet the real-time requirements. The proposed improved model has the characteristics of high accuracy, fast speed, and strong robustness, and is suitable for object detection tasks in autonomous driving vehicles. This study provides a new fusion algorithm for the perception of autonomous driving environments and provides new ideas and methods for the practical application of the algorithm through testing on the Wuhan traffic environment dataset.

In the context of the new era of rapid development of big data and artificial intelligence technologies, the use of data mining and algorithmic techniques to build sales prediction models with high accuracy rates can quickly help cross-border e-commerce (CBEC) enterprises to cope with the challenges related to data processing, feature extraction, and value information acquisition. Through the analysis and mining of raw data of CBEC transactions, a two-layer fusion model is built by using the stacking integrated learning method, with the Random Forest, XGBoost, and LightGBM algorithms as base learners and the support vector regression algorithm as a meta-learner, to predict CBEC sales and compare the prediction properties with the single model constructed by the base learner. The results show that the integrated learner trained by the stacking integrated learning method can improve the prediction performance of the single learner and performs best among all learners. The model can objectively reflect the status of CBEC sales and play an essential role in the company’s execution of strategic decisions such as merchandise sourcing and sales.

In the current deep learning rice disease recognition method, there are problems such as more complex models, poor adaptability to be deployed on edge devices and mobile terminals with limited computational resources, and poor real-time and accurate recognition of crop diseases, so we propose a lightweight rice disease recognition method that improves ShuffleNetV2. Based on the improved Shuffle Block unit called AugShuffle Block, the ECA (efficient channel attention) module is added, and average pooling is used instead of maximum pooling to obtain more feature information, at the same time, the number of Shuffle Block units used in each Stage module in the network structure is reduced to reduce the number of parameters in the model and the amount of computation. Experiments are conducted on the Kaggle rice disease dataset. The results show that the number of parameters of the model is 0.944 × 10⁶, the computation amount is 105.47 × 10⁶ (FLOPs), and the average accuracy of disease recognition reaches 96.02%, provides a reference for the deployment of the rice disease recognition method for applications on resource-constrained devices such as mobile terminals.

The diffusion model has been widely applied in various aspects of artificial intelligence due to its flexible and diverse generative performance. However, there is a lack of research on applying diffusion models in the field of depth map restoration. This is primarily because the diverse generation capabilities are not sufficient to meet the requirements for depth map completion. Depth map completion requires rational completion in the current scene. In order to adapt diffusion models for depth map completion tasks, this paper proposes the Multi Condition Diffusion Model (MCDM). It allows the addition of conditional information to constrain the model’s rational completion of ill-regions. The MultiConditionLN module effectively adds multiple conditions to the depth map completion task. This module uses the completion region mask and the input image as conditions to constrain the model’s generation process. This enables the model to complete the regions that need restoration based on the scene of the input image. The proposed model achieves promising results on depth map datasets.