This PDF file contains the front matter associated with SPIE Proceedings Volume 12635 including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

At present, the evaluation mechanism of higher education is still relatively old, with too much emphasis on theoretical assessment and insufficient assessment on practical application. Because the assessment mechanism is too single, it limits academic freedom and innovation. The inflexible educational mechanism will also lead to the shift of the focus of personnel training. Therefore, we must improve the evaluation mechanism of higher education. In this paper, the posterior probability is selected as an important part of the cost function. The cost function is combined with the Gini coefficient, which is the feature division index of random forests It is to achieve the purpose of considering the sample misclassification cost when the random forests weak classifier selects features. Squaring can increase the cost of misclassifying the actual class as a sample. It can also realize the evaluation of educational resources and verify the experimental results by comparing the performance of a single model. The results prove that this model is more suitable for evaluating the education and training of agricultural talents. It can also measure the application value of resources to maximize the use effect of educational resources.

Aiming at the problem that the traditional algorithm is easy to overfitting, which leads to low prediction accuracy of the model. This paper designs a traffic accident impact range prediction model based on CatBoost ensemble algorithm. The model uses linear fitting for range prediction and uses the ordered boosting method to introduce the prior term and weight coefficient. It can automatically adjust dynamically in each calculation, so as to effectively avoid the condition offset and gradient deviation and reduce the overfitting. Under small-scale training, the algorithm can achieve high accuracy prediction and has strong generalization ability.

The use of power monitoring and management system in power enterprises has significantly improved the convenience, fault tolerance, and related economic income of power grid in power supply. At present, there are still many problems in the management system, such as the lack of unified planning. Many functions related to power monitoring and management often need a lot of software to be repeatedly installed in the system to complete the functions required by customers. Meanwhile, this software often has duplicate functions. Power data cannot be shared, and the upgrade of the system will encounter many obstacles. To meet the data communication requirements of component-based power monitoring and management system, this paper applies the characteristics of multi-point distribution and real-time and efficient transmission of data based on model-based information transmission. It also applies component-based data monitoring technology to complete the design. For the power data content of monitoring and calculation, a reusable framework is applied to facilitate user expansion. Users can complete the interface layout by dragging controls on the control interface of the power monitoring and management system, and then realize data customization and saving by binding the database. Finally, the monitoring of the power grid data platforms under cloud service is realized, which is conducive to the effective and timely processing of blowout power grid data. It is of great significance to the safe and stable operation of the power system.

The purpose of cutting material to length is to cut the material into finished products according to customer needs and minimize the amount of surplus material. To address this problem, we propose a model for extensions to the one-dimensional bin packing problem involving multiple length restrictions, which considers that all finished products have different lengths, and all length requirements are loaded into the finished products to achieve the optimization objectives of minimizing the quantity of the finished material and minimizing the remaining length of the finished material. The model is solved by a hybrid intelligent algorithm consisting of a genetic algorithm integrated with fit algorithms. The effectiveness of the hybrid intelligent algorithm is theoretically validated and experimentally verified using actual data and specific numerical examples.

With the continuous development of intelligent connected vehicle industry, cameras and other vehicle-mounted devices are widely used, so the amount of data collection is increasing. There is a large amount of sensitive information hidden in the image data generated by connected vehicles. Once the data leakage event occurs, it may cause very serious consequences. In order to improve the security of connected vehicle data and reduce the threat of sensitive information leakage in image data, this paper provides a desensitization process of connected vehicle image data, and desensitizes sensitive information based on instance segmentation technology. In this paper, a real road image dataset is collected, and realizes desensitization of the modified dataset based on proposed framework.

This paper relies on the corpus data of UGC (User-Generated Content) and OTA (online tourism) in the tourism market, adopts various natural language processing techniques represented by deep learning, mines the useful information in the corpus, establishes the knowledge graph and mines the association relationship between tourism products by using data mining techniques and association rule algorithms, and calculates the potential associations in each review, each travelogue and each public article related to cultural tourism through Apriori algorithm to mine their potential associations and explain the association phenomenon reasonably, in order to play a role in scientific regulation and commercial market development for government or enterprise.

With the development of science and technology, chemical production technology is more and more advanced and intelligent, but the advanced technology also means the complex process, also inevitably appear complex and difficult to solve the problem, a method by Improved Marine Predators (IMPA) to optimize fault classification of chemical processes by Random Forest (RF) was proposed. First of all, Principal Component Analysis (PCA) was used to extract feature from process fault data, remove redundant data, reduce dimension, and speed up operation. Then, the improved Marine predator algorithm was used to optimize the parameters of random forest, and a fault classification model was built according to the global optimum fitness function. The fault features after PCA dimension reduction were input into the stochastic forest fault classification model optimized by the improved Marine predator algorithm to realize the fault classification of chemical processes. Tennessee-Eastman (TE) process was used to test and compare the constructed PCA-IMPA-RF fault classification model. After verification, it can be seen that the method mentioned in this paper has high accuracy of fault classification and can be effectively applied to chemical process fault classification.

With the increasing amount of flight test data, the traditional single-thread technology has been unable to meet the needs of rapid processing of massive flight test data. In order to improve the efficiency of flight test data processing, the characteristics of flight test data processing are analyzed, and a parallel flight test data processing method is proposed. This method realizes two-level parallel processing of multi-link parallel and data frame analysis parallel of flight test data. The method is verified by actual data, and the results show that the efficiency of this method is significantly improved compared with the traditional single-thread data processing method.

With the rapid development of aviation electromechanical technology, new aircraft are equipped with a large number of sensors and electronic control devices. In order to meet the data transmission requirements of avionics systems, Ethernet Data Distribution (DDS) protocol has gradually become the standard solution for published/subscribed data in distributed real-time systems. Facing the new requirements of bus data processing using DDS protocol, starting from the characteristics and format of Ethernet DDS data recording, this paper gives the overall architecture and core processing method of the processing software. This paper analyzes the key problems of Ethernet DDS data processing and proposes solutions for DDS bus Interface Control Document (ICD) plane structure, multi-level parallel data processing algorithm, data detection and filtering. The method is validated and can correctly process Ethernet DDS data.

The traditional ant colony algorithm is inefficient to search, easy to fall into algorithm stagnation and local optimization problems in UAV path planning. To ensure that the UAV can avoid obstacles and fly safely, the ant colony algorithm is improved and optimized. First, the target planning area of the UAV is modeled in three dimensions using raster method. Secondly, the update rules of pheromones are improved, and the weight factors of pheromones and heuristics are adjusted. An unmanned aerial path planning algorithm based on improved ant colony algorithm is proposed to plan a safe and optimal path for the unmanned aerial vehicle. Finally, the simulation results show that the improved algorithm has a better flight path than the traditional algorithm.

Short-term load forecasting plays a key role in the overall power dispatch and power sales companies' participation in the spot market. This paper proposes a electricity load forecasting method based on the WOA-BiLSTM-Attention Algorithm. It takes historical load data as input and introduces the Attention mechanism to give different weights to the BiLSTM implied states. At the same time, to address the problem of difficult selection of hyperparameters of this model, the WOA algorithm is proposed to realize the optimal selection of hyperparameters of this model. The WOA-BiLSTM-Attention algorithm not only improves the prediction accuracy of the model but also enhances the prediction efficiency and reliability of the model.

In order to improve the safety of Unmanned Surface Vessel (USV) sailing in the ocean with movable obstacles, a new method based on greedy algorithm, ant colony algorithm and grid method is proposed. By adding periodic repeated calculation of movable obstacles with known paths, the route is periodically replanned. The approximation method is used to solve the problem that the traditional greedy ant colony algorithm is difficult to avoid the movable obstacles in the optimal course design of unmanned surface vessel. The algorithm uses greedy algorithm to plan the optimal route and ant colony algorithm to get rid of the situation that greedy algorithm is easy to enter the local convergence state. By rationalizing the movable obstacles on the known path, the navigation water limit of unmanned surface vessel caused by movable obstacles is avoided. The simulation results show that the algorithm is practical and reasonable for optimal route design when USV avoids movable obstacles. The simulation results show that the algorithm has the disadvantage of long search time.

At present, curve and surface fitting is widely used in three-dimensional measurement, industrial design, archaeology, medicine and other fields, and curve and surface fitting has also become a hot spot and a difficulty at present. The surface point cloud data scanned by high-precision 3D laser scanning instruments on site are often complex, and the data are relatively dense for curves. If the approximation fitting is used, complex information may not be reflected enough, and the interpolation fitting may produce over-fitting phenomenon. This paper proposes a feature point selection algorithm, which is more targeted for dense point cloud data than the general cubic B-spline interpolation algorithm. The feature point selection algorithm can retain feature points and remove non-feature points and minimize the number of fitting segments on the premise of meeting the accuracy requirements of the final fitting curve.

With the development of smart homes, more and more smart home devices are equipped with networking features, and networking technology has been widely applied in smart devices. In order to meet the networking requirements of small, low power consumption, low cost and short distance portable terminal equipment, the short distance wireless technology has been developed rapidly. Currently, the short distance wireless networking technologies used in smart homes in the field of whole-house intelligence mainly include star wireless network technologies, represented by WiFi and Bluetooth, and wireless mesh network technologies, represented by Bluetooth (BLE) mesh and ZigBee. Compared to the star wireless network technologies such as WiFi and Bluetooth , the wireless mesh network technologies have the advantage that the device itself can act as a route node for other devices. The communication range is extended by means of multiple hops, then all the nodes can constitute a mesh so that the communication signals can well cover the entire house, avoiding the problem of weak signals or connection failures incurred by the fact that the WiFi device is located in remote corners. However, the BLE mesh and ZigBee wireless mesh network technologies still have limitations. BLE mesh networks use the flooding algorithm to transmit data from the source node to the destination node from near to far layer by layer, which is like flooding. But due to the use of the flooding routing algorithm, all devices need to perform broadcast forwarding in each communication process, resulting in low communication efficiency and high wireless bandwidth occupancy. ZigBee uses a wireless routing algorithm based on an Ad hoc On-Demand Distance Vector (AODV) standard, which is more efficient than BLE mesh, but it is still a bit inadequate in the case of multi-node networks, reflected in not ideal stability and communication delay. In this paper, the O-Mesh mesh routing algorithm with low network overhead and high efficiency was proposed, which effectively improves the performance of the entire wireless mesh network through mechanisms such as central node routing, neighbor node management, multiple node reuse for route discovery, alternate path caching, and central node data forwarding, resulting in a significant increase in total network throughput and average delay. Telink's chip TLSR8258 supporting multiple protocols (ZigBee, BLE mesh) was used for testing and verification, and the results showed that the average communication delay was reduced by 28% compared to ZigBee.

In order to implement precision management on the campus, the decisions need data support, and the crowd density on campus is one of the important parts. Based on campus wireless network logs, which is widely used on the campus, this paper proposes an analysis algorithm to obtain online wireless network user numbers in real time and draws the conclusion that the numbers of online users can represent crowd density on campus. Experimental results show that this algorithm can effectively get the numbers of online users in each area of the campus, and the campus heat map made with these data can reflect the real-time distribution of campus crowd and crowd density. This method uses log analysis method which is a general solution for some problems and has practical value for in-depth analysis.

Backfill scheduling is a common scheduling strategy in high-performance computing systems that allows priority execution of low-priority jobs to make better use of available resources. Job running time is an important parameter that affects the performance of backfill scheduling algorithm. However, in order to avoid job killing due to lack of time, the running time requested by users is often several times higher than the actual running time, resulting in a certain degree of resource waste. In order to improve resource utilization, a new job running time prediction algorithm is proposed by combining classification and ensemble learning methods. The algorithm first classifies the historical job set according to the application type, then uses Jaccard coefficient to calculate the similarity between the jobs, and further classifies the jobs. At the same time, different integration models are constructed for the jobs of different application types. New jobs are categorized, and the class's integration model is used to predict the running time of the new job. The algorithm was tested on the historical job data of the National Supercomputing Center Kunshan, Hefei Advanced Computing Center and "Wuzhen Light" supercomputing Center and compared with GA-sim algorithm and IRPA algorithm. The experimental results show that compared with the IRPA algorithm, the average absolute error of the algorithm is improved by 60% on the three data sets on average. Compared with the GA-sim algorithm, the average prediction accuracy of the algorithm is improved by 20% on the three data sets on average. Through the in-depth analysis of the experimental results, the amplification method for the low estimation of long and short jobs is given.

Aiming at the calculation of the air threat range of the ship route, the paper described the requirements and judges whether the route segment was under threat. On this basis, an algorithm model was established for the specific threat range of the threatened route segment, and a calculation example was given for verification. The analysis and verification results showed that the calculation model proposed in the paper could quickly and accurately calculate the specific threat range of the ship route, it can provide effective auxiliary decision-making reference for commanders or operators when drawing routes.

The adoption of network technology for control and measurement makes Industrial Control Systems (ICSs) and Internet of Things (IoTs) more and more powerful, whose communication data plays an important role to realize data intercommunication and sharing. However, communication data have different characteristics, even if the same communication protocol is used in ICS or IT, the communication characteristics and modes are quite different. In order to understand the difference of communication data between internet and industrial ethernet, in this paper, we conduct a series of experiments that use information entropy algorithm to data changes based on different industrial protocol and http protocol, which always is used to measure the uncertainty and describe the uncertainty of data packets. The experimental result analysis shows that there exists a big difference in industrial ethernet data and internet data, the industrial ethernet data are more relatively balanced and more regular than those of internet data.

Based on the uncertain characteristics of emergency material demand, this paper uses interval number to represent the uncertainty of emergency material demand and considers the two objectives of rescue cost and time to establish an optimization model of emergency logistics facility location and route. The interval constraint theory is used to transform the interval uncertainty into the definite number, and the multi-objective is transformed into a single objective by the non-dimensional linear weighting method. The genetic algorithm is designed to solve the model, and the effectiveness of the model and algorithm is demonstrated by a practical example.

The prediction of traffic accident duration is the basis of highway emergency management. Timely and accurate prediction of traffic accident duration can provide a reliable basis for road guidance and rescue organization. This paper discusses the traffic accident duration prediction method of N-BEATS model in detail. Through the change of sliding window size and the continuous adjustment of the number of iterations, the appropriate parameters are found to produce a good prediction effect. The dataset used in this paper is US Accidents, a nation-wide dataset of traffic accidents covering 49 states in the US. The experimental results show that compared with the classical time series prediction models such as Bi-LSTM, SVM, RNN-GRU and AttnAR, prediction of traffic accident duration model based on N-BEATS proposed in this paper is optimal in the three evaluation indicators of RMSE, MAE and SD, which shows that the model has the highest prediction accuracy and good performance.

The appearance of small microchips has few defect samples, and the effective defect sample data of previous large-packaged chips can support the training of defect detection algorithms to obtain high accuracy. The current mainstream detection algorithms have a poor ability to detect small targets, which can lead to missed detection and false detection. In this paper, based on existing artificial intelligence techniques, such as transfer learning and SSD (Single Shot MultiBox Detector), we propose an intelligent detection system for microchip appearance defects based on transfer learning for the current stage of the chip appearance defect detection problem. In this paper, the appearance defect samples of larger chips in the past are migrated to microchips as knowledge to improve the accuracy of defect detection. The SSD algorithm is improved to enhance the image features, and the idea of cross-level feature fusion is used to improve the feature expression and strengthen the semantic information.

With the development of integrated circuit industry, the complexity of integrated circuit chip design is becoming higher and higher, and the difficulty of verification is increasing exponentially. Verifying that the RTL description is as expected has become an increasingly important part of the modern integrated circuit industry. Compared with the traditional dynamic simulation verification, formal verification belongs to static verification, and it does not need manual inputting excitation, and can achieve the advantages of 100% coverage. This paper mainly introduces the status of formal verification, what formal verification is, the prospect of formal verification in the future, and provides a formal verification tool (VC Formal DPV) application practice model and several practical methods to significantly optimize its performance by taking the divider as an example.

Power electronic equipment’s general fault alarm, based on patrol vector, monitors for anomalies in the data thresholds. When abnormal is found, the alarm shall be timely reported, so as to timely screen the possible abnormalities of the power plant equipment, find out the problems and deal with them in advance. Because most fault cities of stations cause new vectors to deviate from their normal working space, the research in this paper mainly includes data theory, line parameter creation, data modeling, and fault alarm multiplication. First, the data is processed. The data is denoised by wavelet to remove the boundary noise in the data as much as possible and smooth the data. Then, the process-saving moment is established. The police model is constructed, and the process-saving moment is utilized. The experiment verifies that the method in this paper can put forward the included fault information and warn in advance before the fault. It gives the power plant equipment fault warning information, which can remind people to take measures in time to avoid personnel and economic losses.

There are many typical electrical equipment in UHV converter valve hall, especially the UHV converter transformer bushing. Infrared thermal imager shows that the temperature gradient distribution is significant, mainly because of the combined effect of voltage and current with rich frequency. This paper attempts to deeply analyze the voltage and current waveform at each key node of the UHV converter valve hall from the theoretical level, and deduces the analytical formula with a 12 pulse rectifier (double bridge rectifier), obtaining the Fourier decomposition results of the actual voltage waveform of the converter bushing: the DC component is 697.6kV, fundamental frequency is 50Hz, the peak value of the 1-27th harmonic component, THD=46.79%. Furthermore, the nonlinear relationship between the dielectric heating enhancement coefficient and the temperature is obtained by coupling the voltage and current waveform with the converter transformer bushing structure through the theoretical formula. At the same time, the Fourier decomposition component of the voltage at the first 2500Hz of the waveform is intercepted in the calculation of the dielectric joule heating, and the voltage waveform at the first 2500Hz is reconstructed, with good reconstruction results. The research results of this paper can provide a good theoretical guidance and technical support for the operation and maintenance of typical power equipment in UHV converter valve hall.

In recent years, Physical Unclonable Function (PUF) based hardware encryption has raised its significance in information security. Generally, a PUF is evaluated by three metrics: reliability, uniqueness, and uniformity. It is crucial for a PUF to have a good uniformity if it is used to generate cryptographic key. This paper focuses mainly on the Arbiter PUF and proposes a uniformity adjustment scheme for it. The main idea of this scheme is to add an adjustment module to the original circuit and construct an algorithm to automatically set the adjustment signals according to the responses. In the end, we use FPGA to test the uniformity, reliability and uniqueness of the adjusted PUF.

The dynamic pledge financing mode in logistics finance can effectively solve the financing difficulties of SMES, but it has some complexity in process control and risk management. In this study, pledge rate and warning value are selected as the joint control risk indicators of this model. By analogy with the phenomenon of “sharp peak and thick tail” in financial asset return series, GARCH model is introduced into VaR to measure long-term price risk. The effectiveness of the VaR model based on t-distribution and GED-distribution was compared, and ARIMA model was used to predict the discounted value of the pledge, so as to calculate the pledge rate and early-warning value consistent with the risk tolerance of the bank. Thus, the process of accurately measuring the risk index was provided for the relevant parties, and the risk management of dynamic pledge financing was optimized.

In this paper, to predict the nuclear weapons, we first introduce evaluation indicators that affect the possession of nuclear weapons, economic indicators, scientific and technological indicators, and establish a TOPSIS evaluation model improved by the optimal assignment method to predict countries with evaluation values less than 20, as countries that will possess nuclear weapons in the next 100 years. Then, in view of the fact that the number of nuclear weapons is calculated in years and changes over time, and considering the global consensus to limit the number of nuclear weapons from 2022 when the Treaty on the Prohibition of Nuclear Weapons and other policies come into force, it is decided to build a Verhulst prediction model with saturation based on the LS-SVM algorithm, and finally to improve the accuracy and reasonableness of the model by using the metabolic data processing method of equal-dimensional neutrosophic recurrence prediction. By predicting the number of nuclear weapons, countries can make reasonable plans for future nuclear weapons production and hope to reach a global consensus, which will help to solve the nuclear crisis.

Studying the theme of emergencies is of great significance to the emergency management of subsequent public opinion. In order to solve the problem that LDA ignores the context semantics, and the topic distribution is biased towards high-frequency words, Bert is introduced. LDA-Bert public opinion topic mining model is proposed. First, LDA is used to select candidate words; Then, Bert is used to construct candidate word vectors and topic vectors with context semantics, and the topic keywords are filtered twice by cosine similarity calculation; Finally, the corresponding public opinion response strategies are proposed through the subject mining results of different life cycles. Taking the "Xi'an epidemic" as an example, the experiment proved that the model can effectively extract theme keywords, providing a strong basis for the follow-up analysis of theme changes at different stages of public opinion.

As an integral part of our life in modern society, e-transaction under the third-party payment model is widely used. However, this model may cause some issues such as privacy exposure, data being stolen or tampered, and unsatisfactory fund supervision. To solve these problems, the paper proposes a secure transaction system based on the smart contract and the InterPlanetary file system. Blockchain technology is used to realize decentralized transactions and payments, ensure the anonymity of users and stores in the transaction process, and guarantee that transaction records and payment records cannot be tampered with. BF-IBE algorithm is introduced to encrypt transaction records to solve the problem of transaction information exposure and ensure its confidentiality. And InterPlanetary file is used for distributed storage of encrypted orders to improve the storage and query efficiency. Based on smart contract and blockchain technology, traceable supervision of capital flow is conducted to solve the problem of unsatisfactory capital supervision and ensure the correctness and integrity of capital flow. Smart contracts and digital signature technology are combined to authenticate users to ensure the integrity of logistics data, only allowing legitimate users to modify logistics information. Finally, the security proof and experimental simulation and testing are carried out on the new system, the off-chain and on-chain computing efficiency and resource consumption are analyzed in detail, and the results show that the solution is secure and efficient.

Aiming at the application defects of traditional temperature control box in vehicle-mounted scenario, a vehicle-mounted temperature control box based on TEC was developed to control ambient temperature and diagnose hardware faults through UDS protocol. Firstly, the simulation analysis of the temperature control model is carried out, and the fuzzy temperature control algorithm based on the dynamic adjustment of the stretching factor is used to achieve a better response effect than the traditional fuzzy control. Then design temperature detection circuit, CAN communication circuit, H bridge drive circuit and ECU program. Finally, the experimental test shows that the system can communicate with the upper and lower machines through the UDS protocol, realize the temperature control and fault detection functions in the variable theory domain, and achieve the expected effect of the system.

In advanced High-Performance Computing (HPC), convolution operations take a big proportion in convolutional neural networks, and convolutional neural networks very common in image and video based deep learning applications, because of which, this paper takes improving the performance of convolution operation as the research direction. Convolution can be performance in many ways, such as using mathematical definition to calculate, conversing to Fast Fourier Transform (FFT), conversing to batch matrix multiplication (im2col) or using Winograd algorithm. For small filter, Winograd has unique advantages. AMD based ROCm environment, the implementation of Winograd and an optimization method of Winograd based on multi-thread communication algorithm are introduced in this paper. For the Winograd convolution in ROCm 2.9.0, the speed of the algorithm was increased by more than 150% after optimization in this paper. Under some certain computing power situations, the performance of the optimization algorithm approaches or even exceeds cuDNN and MIOpen.

Visual object tracking is a traditional task in computer vision, which has developed with several decades. With the development of machine learning, Correlation Filter (CF) has been proposed with satisfying performance and very high framerate. Though the CF framework has numerous strengths in this task, the tracker is fragile to miss the target in several scenes, including extreme illumination, target occlusion and deformation. Recently, thermal modality, which detects the target’s temperature, is robust to the night scenes and can provide a precise target contour. In this paper, we propose a CF based tracker with decision fusion strategy for visible-thermal (RGB-T) tracking. First, we introduce multi-modal KCF trackers as our baseline. Then, we design a decision fusion method considering the Peak-to-Side Rate (PSR) of the score maps, thereby achieving an adaptive fusing those modalities and avoiding model’s heterogeneity. In the experiments, our tracker has validated on the public dataset, namely GTOT. Compared with two uni-modality trackers, the proposed tracker with real-time speed has shown superior results on both target localization and scale estimation.

Along with the continuous development and deepening reform of the Chinese power market, the main body of the power market trading is increasing day by day, the marketization of the transaction electricity has increased substantially, and the types and cycles of the transactions also become more diversified, which puts forward higher requirements for the settlement of transactions in the Chinese power market. This paper is based on the research of the current development status of Chinese power market transactions settlement, explores the effective operation mode of the power market transactions settlement mechanism, which can promote the establishment of the power market transactions settlement mechanism better and faster.

As a metering device, the electric energy meter is used to detect the electric quantity at each detection point. This paper studies a monitoring method and system, storage medium, and terminal for the verification data of electric energy meters, involving the field of electric power. The main purpose is to improve the existing problem that the operating data in the verification process cannot be grasped in real time, and there is a risk of damaging the pipeline hardware, thus reducing the monitoring efficiency. The process includes analyzing the target verification items of the electric energy meter in the verification work order and determine the monitoring components matching the target verification items. After starting the verification operation of the electric energy meter, send the data acquisition command to the hardware monitoring subunit, and send the data recording command to the software monitoring subunit. If the problems found in the detection cannot be grasped at the first time, the pipeline hardware may be burnt, and other risks may be caused. Obtain the data collected by the hardware monitoring subunit and the operation status data recorded by the software monitoring subunit. Reduced monitoring is a problem of work efficiency. In reality, work efficiency is very important, especially in terms of risks and hidden dangers. Extract the hardware monitoring features of image data and/or infrared signals, and the software monitoring features of operation status data, and monitor the hardware monitoring features and software monitoring features based on the data status monitoring model to determine the monitoring results of verification data.

Aiming at the problem that the information difference between modalities in the current multimodal sentiment analysis model and the insufficient fusion between modalities lead to the low accuracy of network prediction, this paper designs a multimodal sentiment analysis model based on BERT-ResNet50. The model uses BERT and ResNet50 to extract text and image features respectively, fuses multi-modal information through the encoder layer of Transformer, and finally uses the Softmax layer to classify multi-modal information. The dataset used in this paper is the Twitter sarcasm public dataset. Through experiments, the BERT-ResNet50 model proposed in this paper is higher than the comparison models in accuracy, recall rate and F1 value, and the accuracy reaches 74.05%. Ablation experiments show that the accuracy of the model in multi-modal sentiment analysis is higher than that in single-modal sentiment analysis.

In recent years, with the continuous progress of China's high-speed railway construction, railways play an important role in the country's economic development and provide a strong driving force for people to live a better life. The country has put forward higher requirements for train running safety. Track foreign body intrusion is one of the important factors threatening the safe running of trains. In order to ensure the safe operation of the train, the detection of foreign body intrusion on the track has become the top priority. It is necessary to further narrow the detection scope in the actual railway scene graph, and only need to detect whether there is a foreign object invasion in the region of interest. Therefore, the division of the railway area of interest is particularly important. In this paper, the improved Sobel image edge detection algorithm is used for edge detection. The least square method cubic polynomial was used for track line fitting. The track line is translated outward to obtain the region of interest. The division of the region of interest improves the accuracy of the prediction.

Anthocyanin is the main color-forming substance of mango peel, and bagging is an important technical measure to regulate the formation of fruit apparent color. In this paper, using 'Renong1' and 'Ono' as materials, combined with computer image recognition and liquid chromatography technology, the differences and accumulation characteristics of anthocyanins in the peel before and after bagging were compared. Using computer image recognition technology, the skin color of different mango varieties before and after bagging can be quickly screened and identified. Ten indicators such as second-order moment and third-order moment are used for feature extraction, and then the principal component analysis of the indicators is carried out through MATLAB, and the results are obtained. The results showed that five major anthocyanins could be identified in the peels of two different mango cultivars before and after bagging. For the two mango varieties, the accumulation patterns of anthocyanins after bagging were exactly opposite. Bagging of the light-dependent variety ‘Renong1’ inhibited the accumulation of anthocyanins in the peel, while ‘Ono’ accumulated anthocyanins after bagging, showing pigmentation independent of light. This indicates that the bagging technique can be used to breed new red mango varieties with more easily colored skins. This study provides a theoretical basis for the cultivation of new mango varieties, color metabolism and regulation research, and has important application value. At the same time, combined with computer image recognition technology, it provides a fast and efficient detection method for the preliminary screening of plant anthocyanin components.

In artificial intelligence, virtual environment and reality technology are one of the most popular types of research. The integration of graphics, images and human action recognition technology is the focus of improving training methods in sports training. Therefore, the research combines dynamic time warping algorithm and Local image feature action recognition combined with virtual technology is applied to sports training to improve training effects and competition results. The results show that the overall accuracy of the research model is 20% higher than that of the traditional gesture recognition system, and the action recognition accuracy of the self-collected data set reaches more than 90%. In the empirical analysis, the posture recognition system has a good anti-interference ability with a recognition accuracy of about 95% for different parts of the body. Athletes' overall satisfaction with the system reaches more than 90 points, effectively improving the training effect and competition performance.

This paper proposes a fall detection technology based on the YOLOv5s algorithm to solve the problem of hit injury. The method is designed based on the embedded ARM development board of Orange Pi Zero 2. The camera is used to collect human data in real-time, and algorithms train the collected data and are finally verified. The experimental results show that: (1) this method has a reasonable success rate of recognition for standing, walking, and falling, but the success rate of recognition for squatting needs to be improved; (2) Compared with the OpenPose algorithm, the YOLOv5 algorithm has better accuracy, precision, and average accuracy means, but the performance in recall rate is not very good.

Due to the high cost, complex construction, and immature algorithm of existing force feedback equipment, the majority of it is utilized in precise industries such as telemedicine and aerospace. Nevertheless, specific industrial production tasks require force feedback devices immediately. In order to make force feedback equipment extensively employed in the field of industrial robots, a spring-based variable force machine manpower feedback control system is developed. The force state of the robot in the working environment is communicated back to the manipulator via spring deformation, allowing the manipulator to intuitively sense the force environment in the working environment, hence facilitating robot control. The experiment demonstrates that the force feedback control system based on this method has a stable feedback effect and a simple structure, as well as the ability to realize the feedback of varied forces based on diverse application circumstances, which has a high promotional and practical value.

The probability of loan default is one of the most important activities in the financial sector. In this context, lenders issue loans to borrowers in exchange for a promise to repay the principal and interest. In this paper, we use a Bayesian deep learning model to build a predictive model for high performance loan default probability. In the practical case of loan default modeling, we cannot use clean and complete data. Some of the potential problems we inevitably encounter are missing values, incomplete categorical data and irrelevant features, thus requiring data pre-processing. In this paper, we train our model by analyzing the Kaggle Lending Club loan dataset from 2007 to the third quarter of 2017. The results show that our model has more than 96% accuracy. Compared with popular classification models, our model has higher performance.

In recent years, with the rapid development of computers, the Internet and other modern technologies, the information age has quietly arrived. Libraries, as an important way to learn knowledge, have thus ushered in great development. In response to the problem of traditional libraries' static web page compatibility, this paper designs how a responsive library platform based on SSM framework and Bootstrap can be deployed under Centos Linux platform, tested and packaged through Docker technology. Docker covers all stages of platform development and operation and maintenance, and with the help of Docker technology, it can be built once and Docker covers all the phases of platform development and operation and maintenance, and it can be built once and run everywhere with Docker technology. When using LVS to deploy Web server clusters, Docker images can be used to deploy Web servers directly. The test results show that the platform has strong portability, saves the development and deployment costs, and has great advantages in the rapid deployment of large-scale applications.

Secure Multi-Party Computation (SMPC) has attracted the attention of many scholars in recent years because it can calculate an agreed function without a trusted third party. As a specific problem of SMPC, Multi-Party Private Set Intersection (MPSI) is what we focus on. MPSI allows multiple parties to calculate the intersection of their set data securely without leaking any other private data. Although many MPSI protocols have been proposed, they are still not efficient enough. To solve this problem, we propose an efficient MPSI protocol which is based on Bloom filter and ElGamal cryptography. By using encrypted Bloom filter and multiply homomorphism of ElGamal cryptography, our protocol has the following advantages: (1) each party has his own public key and secret key, which increases security; (2) with the same number of parties and set elements, our protocol is highly efficient compared to the related protocols.

Graph neural networks have been widely used in many fields. Most studies are used to solve the node classification task, but most of the time the distribution of the data is unbalanced, which affects the classification accuracy of the model. In this paper, we balance the data distribution in the graph by generating new samples in the embedded space and introducing random variables to control the spatial distance between the new samples and the target samples. We also propose a framework to solve the unbalanced node classification task in the graph, and the experimental results show that the method is effective.

Wireless sensor networks are generally deployed in remote areas and areas with complex geographical environments. It is difficult to replace sensor node batteries. Energy acquisition sensor network nodes are powered by solar cells to provide energy to wireless sensor nodes. The randomness and uncertainty of solar energy in the monitoring area make it impossible to continuously provide energy for sensor network nodes. By predicting the results of solar energy and combining the information collected by the wireless sensor network, the energy usage of wireless sensor network nodes is reasonably planned, thereby improving wireless transmission. sensor network lifetime and the accuracy and reliability of sensor node measurement information. Prediction of solar energy in the monitoring area is an important part of improving the monitoring quality and life of wireless sensor networks. In this paper, the BP neural network combined with the climatic factors in the wireless sensor network monitoring area, such as illumination, the average diffuse reflection intensity of the solar panel on the day, etc., are used as reference data to predict the solar energy in the wireless sensor network monitoring area. Compared with the traditional Exponentially Weighted Moving Average algorithm (EWMA), the error rate of the prediction results is lower, and the prediction effect is better due to the comprehensive consideration of various climatic factors.

In recent years, deep Convolutional Neural Networks (ConvNets) have achieved great success in many vision tasks. While larger and deeper networks perform better, deploying these models on resource-constrained devices makes it more challenging. CNN models can be effectively compressed and accelerated via filter pruning. The traditional filter pruning method is based on a pre-trained model, pruning, and then recovering model accuracy through fine-tuning. However, recent research had shown that using pre-trained models to prune a model reduces the optimization space of the pruned model. Obtaining a pre-trained model is also computationally intensive. A further weakness of the prior works is that the process for selecting filters is too simplistic and insufficient for identifying possibly more important filters. To address these weaknesses, we propose a pruning method that does not rely on a pre-trained model and instead uses the information entropy of the channels during training as a criterion of the filter’s importance. In each training step, we calculate the importance of each channel in each layer of the network based on information entropy. Since the channel in the current layer is calculated based on the filter in the previous layer, the importance of the filter can be equated with the importance of the channel. Then the least important filters are pruned according to the pre-defined pruning rate, and the next training step is resumed. Note that to guarantee the capacity of the pruned model, we still allow those pruned filters to be updated in the next training step. It is experimentally demonstrated that our proposed method outperforms the traditional weight-based magnitude pruning method largely on the CIFAR10 and CIFAR100 datasets using the ResNet series networks and the VGG16 network. On the ResNet56 network, there is only a 0.1% accuracy loss at 40% pruning using our pruning method on the CIFAR10 training set.

Face spoofing detection techniques have performed well in the digital and physical domains separately. However, existing methods do not work well when both types of spoofing attacks need to be resisted at the same time. We propose a new polarization-based unified spoofing detection method for cross-domain face spoofing attacks. With our cross-domain unified spoofing detection framework, our methods can automatically detect and identify face spoofing attacks in both digital and physical domains. In addition, we build a new face anti-spoofing dataset containing polarized modality. We first provide a method for generating polarimetric face images from visible images, which are used to provide a digital domain spoofing attack. Then, we fake faces through physical methods such as photo and mask. In our new dataset, extensive experiments show that our method has better performance and robustness in face cross-domain attack detection and can still defend against cross-domain face attacks with a very small training data size.

Recent work has shown various interesting semantic image manipulation methods based on GAN guided by text descriptions. A method based on GAN inversion can achieve versatile image manipulation functions without a time-consuming preprocessing stage. However, the method suffers from a lack of self-adaptation due to the intrinsic conflict between multi-objective losses. Meanwhile, the method applied in image manipulation guided by text conditions is not robust due to the vast and ambiguous search space. To solve the above problems, we propose a novel framework RAIN based on GAN inversion, which can achieve robust and adaptive text-driven image manipulation. As shown in Fig. 1(c), RAIN contains two main parts: CEV Initialization and RAGAN inversion. CEV Initialization can adaptively provide a Candidate Editing Vector (CEV) in a short time. RGAN inversion is a multi-stage optimization scheme utilizing the CEV as prior knowledge to prune search space. In RAGAN inversion, we explore how to improve the vision-language model's perception capability to restrict search space further. The objective of the paper is guaranteeing semantic correctness and image quality in a time-constrained scenario compared to the SOTA image manipulation methods guided by text descriptions. Extensive experiments show that RAIN can manipulate images guided by text description while meeting robustness and self-adaptation.

In order to provide a set of quantitative evaluation of the comfort of hand-held sports products, this study preliminarily constructed the comfort evaluation index system of hand-held sports products from the perspectives of performance, tactile characteristics, quality and appearance, physical stimulation, discomfort, morphological characteristics, user demographic characteristics and hand characteristics. Secondly, 165 groups of data were collected through the experiment, and factor analysis was used to reduce the dimension of the original index. Then neural network was used to establish the mapping relationship between the evaluation index and the comfort evaluation, and the evaluation model of the comfort of hand-held sports products was obtained. The results show that the relative error between the predicted value and the actual value of the network model is basically less than 5% in the practical application of the evaluation method taking the badminton racket as an example, which reflects that the comfort evaluation method can effectively reduce the correlation between the original indexes, achieve the accurate evaluation of the comfort of products, and verify the effectiveness and feasibility of the evaluation method. The combination of neural network and factor analysis can quantitatively analyze and evaluate the comfort of hand-held sports products and provide reliable suggestions for manufacturers to design and produce more comfortable products in the form of data.

To improve the detection effect of image steganalysis on whether unknown images contain secret information, a novel deep steganalysis model based on coordination attention mechanism is proposed. Firstly, a high-pass filter groups is used to process the input image to acquire the noise residual features almost unrelated to the image content to eliminate the effect of the image content, which is conducive to the subsequent convolutional layer to model the image features. Secondly, the convolutional layer and attention mechanism are combined in the body part of the network, which can not only enhance the useful feature flow, but also boost the network to consider the discriminating features. Finally, the global average pooling layer is employed to aggregate features of channel dimension to generate the feature descriptors, and the fully connected layer is utilized to classification. The probability of input image being identified as the stego image and the cover image is acquired by the softmax operation. A lot of experiments prove that our network is better than the existing deep image steganalysis model.

Aiming at the problems that the target detection accuracy of inland watercraft is reduced, and the difficulty of labeling samples is increased due to fog, according to the characteristics of inland water ship detection, an adversarial unsupervised domain adaptive method is proposed, using labeled sunny samples and unlabeled foggy samples, which can adaptively detect ship targets in foggy conditions. On the basis of Faster-RCNN, a ship local feature domain discriminator based on full convolution and a ship global feature domain discriminator based on a hybrid attention mechanism are respectively constructed to realize the features of sunny and foggy ship images in multi-level domain space alignment, to better complete the adaptive detection from sunny ship data to foggy ship data. At the same time, using the idea of adversarial learning, a local feature adversarial loss function based on the least square method and a global feature adversarial loss based on cross entropy are designed to strengthen the adversarial training between feature extractor and domain discriminator. The experimental results on the inland ship data set show that only using the marked sunny ship data and the unmarked foggy ship data can adaptively detect ships in foggy weather, and the accuracy and real-time performance are significantly higher than those of the comparison model.

Traditional rumor detection methods that only focus on text content have achieved certain results. However, with the rapid development of social platforms, graphic information has occupied a large proportion. In this scenario, traditional detection methods cannot make full use of picture information for rumor detection. Aiming at the above scenarios, a rumor detection model integrating multi-modal features is proposed. Firstly, text features and visual features as well as their hidden states are extracted by using the pre-trained deep learning model, and then the preliminary fusion features are obtained by integrating the hidden states of text and image through the attention mechanism. Then, the text features, preliminary fusion features and social features are spliced, and the image features, preliminary fusion features and social features are spliced to obtain two final fusion features. Then the two features are input into different full connection layers to get their respective prediction results. Finally, the two prediction results are integrated to obtain the final detection results. Experimental results show that the proposed model is effective in detecting multimodal rumor data.

Compared with foreign countries, there is less research on the transfer of text styles in China. Jin Yong is the most famous martial arts novelist in China. So, imitating and learning the styles and methods presented in his works has a significant impact on the growth of the field of martial arts novels in China. This paper conducted an experiment on the transfer of text style in 14 novels by Jin Yong. Based on the transformer model, the experiment used the sentences obtained from Jin Yong's novels through translating back and forth, cleaning and word segmentation to adjust the training corpus and test corpus as the model, adopted the pre-training and fine-tuning method to adjust the model, and used the BLEU4 value as the evaluation index. The experimental results demonstrate that the baseline model performs well overall, and the generated text has a certain style of Jin Yong's novels. After pre-training and fine-tuning, the BLEU4 value of the model is further improved.

In order to better measure the correlation between entities and solve the problem of link prediction in the knowledge graph, a knowledge graph completion method based on graph neural network and attention mechanism is proposed HRGAT. The hybrid attention mechanism is used to calculate the correlation of adjacent entities in the knowledge graph and obtain the corresponding feature entity vector. The capsule network is used as the decoder to better represent each attribute dimension of the entity, to improve the expression ability of migrated feature entity vectors, link prediction experiments were conducted in the classic datasets WN18RR and FB15K-237. The experimental results show that compared with other classical algorithms, this method has some improvement in four common evaluation indicators. It is proved that the hybrid attention mechanism can effectively improve the prediction accuracy of RGCN on the knowledge graph triplet, thus expanding the knowledge map and providing data support for downstream tasks.

In the future, various factors affecting user satisfaction will be analyzed through the survey results to provide the basis for decision-making, to achieve earlier and more comprehensive improvement in user satisfaction. This paper constructs a random forest model to explore the influencing factors of mobile user scoring. Foremost, preprocess the acquired data. Different data are processed differently. The continuous variables are inserted by mean value. Delete obvious abnormal samples. The string variable uses mode to fill in or delete the sample, which is explained in detail in the text. Label coding and unique coding are carried out for the category feature data, feature construction is carried out, and the information field of opinion type (non-stereotyped) is deleted. A random forest classification model is established for the processed data to predict the sample score. In order to make the model more optimized, the Bayesian parameter adjustment method is used to make the model achieve the optimal effect. The results show that the random forest algorithm plays an important role in user rating prediction and will play an important role in service improvement and guiding the development direction in the future.

In recent years, deep learning has excellent applications in many fields, especially in computer vision. However, it has been found that deep learning systems are vulnerable to subtle interference, which is almost imperceptible to humans. This leads to model decision errors and potentially disastrous consequences. Therefore, AI scientists have carried out much research on this, and MagNet, Feature Squeezing and Defense-GAN are excellent research achievements. In this paper, we propose to improve MagNet. By proposing a new network structure and integrating ensemble learning technology, experiments show that this MagNet combined with ensemble learning technology has many advantages over the original MagNet. From the perspective of training time, the training time required by MagNet based on ensemble learning is reduced by half compared with the original MagNet. From the perspective of defense capability, Under the condition that the training time is sufficient and long, there is little difference between their defense capabilities. Under the condition that the training time is limited, MagNet based on ensemble learning is much better than the original MagNet. Similarly, according to this experimental idea, this paper conducts comparative experiments with Feature Squeezing algorithm and Defense-GAN algorithm from the perspective of training time and defense capability and finds that MagNet based on ensemble learning also has advantages.