This PDF file contains the front matter associated with SPIE Proceedings Volume 12700 including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

This paper proposes a distributed GRS code system based on cooperative communication and channel coding. Taking advantage of the fact that GRS codes are MDS codes, a relay information selection algorithm is proposed to optimize the relay information selection method, so as to obtain a relatively better weight distribution of the codeword set at the destination point. In addition, based on the channel coding characteristics, two joint decoding algorithms are proposed at the destination, namely NAIVE decoding algorithm and SMART decoding algorithm. The performance improvement of the relay information selection algorithm on the distributed GRS code system is verified by Monte-Carlo simulations, and the performance of the two joint decoding algorithms is compared.

Connection is the effect between events. It is an important basis for judging of violations and source-trace of failure in cloud service accountability. This paper presents a survey on the event connections including the features, method of discovery and correction. Most of recent research on event connections in cloud services relying on manually labeling and correcting to event connections, and a few of them have achieved automatically labeling of event connections. Finally, we give out a taxonomy of the recent research, some problems in the method of discovery and correction, and an approach in the paper.

Aiming at the problem that the carbon emissions of enterprises change with the change of production mode, which leads to the deviation of enterprises in the process of carbon emissions prediction and affects the accuracy of enterprise carbon emissions prediction, an enterprise carbon emission prediction algorithm based on multi-dimensional feature data is proposed. Analyze the influencing factors of carbon emissions of enterprises, collect multi-dimensional data of the influencing factors of carbon emissions through automatic collection and manual filling, and pre-process them. The principal component analysis and kernel function algorithm are used to extract multidimensional data features, and the extracted multidimensional features are transferred to the Drosophila - differential evolution optimization RBF neural network model, and the best enterprise carbon emission prediction results are obtained through continuous learning of the model. The experimental results show that the proposed algorithm can effectively predict the total carbon emissions of enterprises in high and low emission scenarios each month, accurately predict the carbon emissions during production, processing and combustion, and predict the emissions of atmospheric pollutants such as nitrogen oxides and inhalable particulates in the process of carbon emissions of enterprises, with good prediction effect.

Internet of Things (IoT) enables traditional systems to evolve into a new form of interaction and integration among people, machines, and things. Process-Aware Information Systems (PAIS), supported by Business Process Management (BPM), can provide programming paradigms for IoT's business processes. However, the heterogeneity and continuity of IoT data make it difficult for traditional PAIS to handle continuous events. One of the keys to solving this problem is to encapsulate the processing logic of heterogeneous, streaming IoT data from different sources in a unified manner, generate business events of interest to the business process, and integrate them into the business process. An event-publish/subscribe mechanism-based IoTService model can integrate IoT events into business processes and enable PAIS to respond to physical world changes. Case studies and experimental research have proven this model's effectiveness and efficiency.

Community search is a widely used technique in graph data mining that aims to find communities containing a given query node. While existing works have mainly focused on homogeneous information networks, most real-world networks are heterogeneous. To address this, this paper proposes a weighted k-core community search method designed for heterogeneous information networks. Firstly, the influence of the association weight between nodes based on meta-paths on the community search results is considered, and a weighted k-core community model (k, P)-Wcore is established, thereby improving the accuracy of community search. Subsequently, in order to improve search efficiency, an optimization algorithm OptWcore based on graph traversal search space is designed. This algorithm can effectively reduce redundant calculations and reduce the depth of path search, thereby improving search efficiency. Finally, experiments conducted on four real-world heterogeneous information network datasets demonstrate the effectiveness and efficiency of the proposed method.

In the transmission network of automatic hydro-meteorological observation system of ocean stations, the observation data are transmitted to In the transmission network of automatic hydro-meteorological observation system of ocean stations, the observation data are transmitted to the data center in clear text, and the typical communication methods are 3G/4G/5G wireless communication, BeiDou-2 satellite navigation system, and In order to solve the problems of incomplete real-time observation data caused by communication failure and easy leakage of clear text data transmission, a double-encrypted data transmission device with multiple concurrent communication methods was designed. the design of the hardware circuit, software, double encryption algorithm, system implementation and demonstration applications were presented. design result was applied by demonstration, and the application result shows that the parallel transmission of multiple encrypted data can ensure the system design result was applied by demonstration, and the application result shows that the parallel transmission of multiple encrypted data can ensure the complete and secure transmission of real-time data to the data center, and the functional performance index meets the requirements of operational business.

Aspect-level sentiment classification aims to determine the sentiment polarity of a given target aspect in a sentence. To solve the problem of ignored the impact of noise in sentences and sentiment relations between different aspects on the sentiment classification performance of models in the current studies, this paper proposes an aspect-oriented syntactic dependency graph and an inter-aspect dependency tree. Based on it, an interactive graph attention network model is proposed to extract sentiment features of the target aspect by exploiting aspect-oriented and inter-aspect information. Experimental results on SemEval-2014 and Twiter datasets show that the sentiment classification ability of the model is superior to the baseline models, and the accuracy of sentiment classification on restaurant review dataset (Rest14) reach 83.36%.

The statistics on demand of customer is very important for business, such as collecting location information of customers to provide them with various services. In this paper, we present a new method for customers to authenticate their identity and position without revealing them and statistics the distribution of passenger flow in different periods, where the purpose of privacy-preserving could be achieved. Due to the identify of customers changes rapidly, in order to ensure that the identify of customers can be updated in time, this paper use dynamic group signature method to add and delete identify frequently. In the meantime, customers need to be divided into different types to meet the needs of different business. Therefore, we will use customers' age and occupation as their identify marks. In order to understand the needs of customers of different ID, age and occupations that we statistics the identity to help business find the suitable development areas.

Schumann resonance is a phenomenon in the extremely low frequency electromagnetic field, which is generally presented in the form of spectrum diagram and time-frequency diagram in the form of data. In this paper, based on the analysis of the data of Schumann resonance 0~50Hz time-frequency diagram, the time difference matrix and frequency difference matrix of the first and second order of the Schumann resonance time-frequency diagram are extracted respectively, and the time-frequency diagram is encoded and decoded by convolution autoencoder. In the process of encoding and decoding, the mapping function is constructed by combining the constraints, and the time-frequency diagram is decomposed and fitted, so as to improve the ability of data to reflect changes. Then, according to the seven different index matrices obtained, the determinant, rank and trace of each index matrix are obtained, and the earthquakes with magnitude above six in the northern state of New Zealand from 2016 to 2019 are analyzed. The final experimental results show that there is a certain degree of feedback phenomenon in the selected analysis indicators in the research object of this paper for a period of time before and after the event, which shows that the method of time-frequency diagram analysis is feasible.

Semantic role labeling serves as a crucial approach to obtaining semantic information and enabling shallow semantic analysis. Currently, the BiLSTM-CRF model is the primary method used for semantic role labeling. However, this model has many network parameters and is unable to effectively capture semantic information in long sentences. To address these issues, this paper proposes a CNN-BiLSTM-MaxPool-CRF fusion model for Chinese semantic role labeling. The model utilizes MaxPool to sample and extract the output of the BiLSTM network to optimize the network structure. Convolution kernels of differing sizes are utilized to capture local features of sentences, and these features are then combined through average pooling to form new feature vectors. These new features incorporate semantic information of the sentence context and are inputted into the model alongside multi-level linguistic feature groups such as part of speech and sentence phrase structure. Through multiple sets of experimental demonstrations, the method proposed in this paper has demonstrated significant improvements in the performance of the semantic role labeling model.

Aiming at the requirements of data query, monitoring, analysis and display during system operation, a real-time distributed data collection and analysis system is designed. This paper first analyzes the processing process of data acquisition and analysis, on this basis, combined with the actual application requirements, minimizes the changes to the existing equipment environment, adds intelligent acquisition cards at the device interface for signal-level or business-level data collection, and adopts cloud/server technology and MQTT lightweight communication technology to visualize the system operation process, and supports on-demand subscription to ensure that problems in the process of system operation can be found in a timely and effective manner.

Due to the Internet's tremendous expansion in recent years, quite a few students have started studying on massive open online course platforms. The information explosion of online education platforms makes it challenging for users to choose their courses effectively. The course recommendation system has become the most effective way for online education platforms to solve the above problems. One of the classic algorithms used in recommendation algorithms is collaborative filtering, but it also has many limitations: 1) Collaborative filtering is affected by cold start and sparsity easily; 2) Collaborative filtering needs to be used for rating information on courses, but the real user data in the online education platform basically does have user rating information on courses nearly, so it is challenging to apply collaborative filtering in the field of course recommendation. So, we propose an algorithm and define a formula to calculate the user's degree of interest (rating) to the course based on the user's interaction information in this paper. It adopts the knowledge graph embedding representation learning method to embed the semantic information of the course into the low-dimensional semantic space, and then use the similarity calculation formula to get the acquaintance between users according to the embedded expression and perform collaborative filtering calculation. Our method is compared with user-based collaborative filtering and item-based collaborative filtering and the model is effective on real data, according to experimental results.

Using the data of automatic weather observation station, intelligent grid forecast, forest vegetation underlying surface, this paper selects meteorological factors such as temperature, precipitation, wind speed, relative humidity as dynamic factors, and the vegetation underlying surface data matching the intelligent grid based on geographic information as static factors. A 1km×1km refined forest fire weather grade forecast model based on data superposition was built, and GIS spatial analysis and visualization technology was applied to the construction of forest fire risk model and the production of fire risk meteorological grade forecast graphics. The business application of this research is of great significance to the establishment of a complete forest fire prevention system in Dalian, the protection of the natural environment of the forest, the maintenance of sustainable ecological development and the safety of human life and property.

The existing intelligent transportation system for device management and control may have some challenges, such as simplistic traffic device management, difficult traffic data storage and management, and difficult to guarantee traffic data security. These challenges all bring adverse effects to the smooth development of the work of the traffic department. To address the above problems, this paper designs a blockchain-based intelligent transportation device and data management method. The information of the devices connected to the system is anchored to the blockchain through the blockchain deposition technology, and the whole life cycle of the devices is managed. The data collection and distributed storage of the traffic devices are realized based on the smart contract technology. Combined with the tamper-evident and traceable characteristics of blockchain, this method provides scientific management of device and device data with high security and scalability, and provides ideas and methods for building a security management architecture for transportation device control system.

Aiming at the problem of safe and stable operation of the new energy site test vehicle, a new energy site test vehicle motor battery data acquisition and monitoring system based on CAN is designed. After the system is powered on, CAN bus is used to collect the running data of the motor and battery in real time, and the sliding average method is used to preprocess the data. In addition, the data is sent to the OneNET cloud platform and mobile APP through WiFi for data visualization. The experiment proves that the system can realize real-time collection and monitoring of the motor battery running data of the field test vehicle, achieve the purpose of equipment safety and stability, man-machine cooperation, and provide an important basis for the test vehicle power system optimization and matching.

In view of the characteristics of multiple test types and evaluation indexes of information equipment performance, this paper studies the reliability measurement of prior information from different sources of test information, such as physical test, semi-physical test, simulation test and physical substitution test, starting from the distribution difference of test data, and carries out the research on the fusion evaluation method of numerical and real tests based on prior information, to improve the test efficiency, shorten the development period, reduce the test cost, and provide technical and theoretical support for obtaining more reliable equipment test evaluation results.

As a new type of energy, wind power and photovoltaic power generation are clean, low-carbon and renewable. As a reserve resource of modern power energy, compared with traditional fossil energy, they have their natural advantages. Since the beginning of the 21st century, scenic power generation has maintained a rapid development trend in the world, and its proportion in the power energy continues to grow. It has become a part of the modern power system that cannot be ignored. Compared with traditional fossil energy, solar power generation itself has great intermittency and uncertainty, which is also the main reason limiting the efficient use of solar power generation at present. In order to reduce the impact of large-scale grid-connection of solar power generation on the safe and economic operation of the power grid, it is necessary to predict the power of wind power and photovoltaic power generation. Accurate power generation prediction can reduce the risk of power grid operation and improve the economy of the power grid.

Financial time series usually consist of multiple time series, and financial time series data forecasting models use the historical data plays of multiple driving series to predict the future values of the target series. In recent years, attention-based Long and Short-Term Memory (LSTM) neural networks and Temporal Convolutional Networks (TCN) have been widely used in time series forecasting. In this paper, we propose a two-stage attention-based TCN and LSTM hybrid forecasting model, in order to better obtain the spatial correlation of driving sequences, we used causal self-attention to obtain the spatial attention weights of driving sequences, then use TCN to extract the short-term features of the series in the first stage, in the second stage, adding the temporal attention module computes the sequence adaptively assigning weights to the input sequence for the current and historical moments, and finally use LSTM to capture the long-term dependence of the time-series data. We used the NASDAQ 100 stock dataset and the financial time series of CSI 300 companies to measure the performance of the proposed model in financial data forecasting.

Passwords are critical issues in the world of cyber security. Unfortunately, despite best efforts, passwords continue to be compromised and leaked onto the Internet, leading to an alarming number of compromised passwords in circulation. In this study, we compare honeypot-captured data from 2021 and 2023 to measure the prevalence of compromised passwords in real-world cyberattacks. Specially, we designed and deployed an online SSH honeypot on the cloud server to capture the latest cyber intelligence in the wild. Our findings show that over 90% of brute force attacks involve the use of compromised passwords, indicating a high level of password vulnerability. Additionally, we observe that the effectiveness of strong-password policies in mitigating such attacks appears limited. This study highlights the need for better password security strategies to counter the high prevalence of compromised passwords in cyberattacks.

In today’s computing era, lots of files are generated from various areas due to the rapid development of technologies. Storing and processing massive, small files is one of the significant challenges for Ceph. Ceph is a scalable, reliable, high-performance storage solution widely used in cloud computing. However, for a large number of small files, Ceph has problems such as write amplification will cause performance bottlenecks. This paper proposes a novel technique Extended Small Files Processing Framework (ESFPF). Firstly, for efficient storage of files, the small files are merged after deduplication, which will effectively reduce the data blocks of Ceph to reduce load to achieve high-efficiency data processing operation. Secondly, a prefetching mechanism and file index is introduced to improve the efficiency of accessing small files. The experimental results indicate that the proposed approach can improve the efficiency of storing and accessing massive, small files on Ceph.

Due to the sampling and difficult quantification of light itself, people often cannot have an accurate understanding and evaluation of light pollution. Therefore, based on the current situation and harm of modern light pollution, this paper analyzes the factors causing light pollution from multiple perspectives, applies the multiple linear evaluation method and uses the least square method to estimate the parameters, and further uses the entropy weight method to determine the weight of relevant parameters, so as to construct a kind of light pollution risk level evaluation model. Four regions with different development levels were selected for numerical simulation to verify the correctness of the theoretical analysis of the model. The calculation of light pollution indicators through the model can help people more comprehensively understand and evaluate the light pollution level of a region or city.

With the continuous expansion of distribution network scale, the real-time performance of distribution network segmentation algorithm is increasingly difficult to meet the requirements of practical engineering. The result of network segmentation is actually to divide a whole into various smaller modules and calculate and solve these small modules respectively. Based on this, this paper adopts a new network segmentation method integrating branch cutting and node splitting. In this way, nodes and branches are operated accordingly. The boundary variables generated by this way include both the voltage on the split node and the current on the cut branch. Using a new model of distribution network - hierarchical topology model, the global optimization problem is transformed into a multi-stage segmentation problem. At the same time, the negative fast division of distribution network is realized by searching for split nodes and cutting branches. The simulation results of an example show that the algorithm is correct, effective and fast.

At the beginning of the formation of the online data trading market, the value recognition of a data product and the transaction are often inseparable from the online game process between the buyers and the sellers. This research focuses on the study of the online efficient pricing mechanism of the whole life cycle of the data product, and on this basis focuses on the formation mechanism of two types of prices. First, it involves the internal benefit distribution of the supply side, and the production and circulation of data products are jointly completed by multiple parties. In this paper, the cooperative game model which of data resource providers, data derivative processors and data exchange platforms is built, and the benefit distribution mechanism based on Shapley value method is designed. The calculation proves the rationality and feasibility of profit distribution. The second is the determination of the transaction price of data products on the supply side and the demand side. This paper designs a set of two-stage online bargaining mechanism. The first stage is the bargaining of the value dimension factor, the second stage is the bargaining of pure price, and both stages can set multiple rounds of dynamic game according to the actual situation of the transaction.

Power demand response is a significant method to provide regulation resources for the power system by adjusting the operating power of equipment. To obtain the regulation potential of each equipment, the load disaggregation is necessary from a large number of the smart meter data. However, the existing disaggregation methods mainly focus on residential and commercial users with the similar and fixed equipment. The industrial user loads are various, which is affected by many factors, such as the climate. To solve the above problems, this paper proposes a load disaggregation method for industrial loads considering economy-climate-production factors based on clustering technologies. First, the main affecting factors of the industrial production are analyzed from three dimensions, including the production load, the non-production load and the security load. On this basis, data marking and clustering methods are proposed to classify the industrial load data. Subsequently, an industrial load disaggregation method is presented for obtaining each type of the load power based on the single variable method. Finally, the effectiveness of the proposed models and methods is illustrated by numerical studies.

In order to actively respond to the requirements of green development and seek new breakthroughs in the whole process and cycle of green transformation such as industrial chain and management links, this paper introduces the time series algorithm into the research, mines and analyzes 141076 historical data collected by the data acquisition system and takes the number of diners as the modeling goal. The empirical results show that the ARIMA(6,1,4) model established in this paper has good analysis effect, and the error rates of the model in the training set and test set are 13.53% and 10.182%, respectively, which can provide a good theoretical basis for achieving precision and saving in dining supply and reducing food consumption.

A problem of using the user review for scoring restaurants from Yelp’s dataset of restaurant business is discussed. Reviews posted by users on the same product or service are diverse and subjective. The sentiment and focus of user reviews are more likely to be subjective even for the same product and service. A review rating model with subjective tendencies of users is constructed through using sentiment classification and cluster analysis to analyze the subjective vocabularies and sentiment coefficients in reviews. The model identifies the terms of different categories in user reviews, quantifies and analyzes them, combines the sentiment with categories, and finally selects the rating of the restaurant as the dependent variable and the elements including the food quality, the restaurant ambience, the service and the grade of recommendation as independent variables before constructing user ratings by using a traditional least squares multiple linear regression model.

A low-frequency phase meter based on STM32 single-chip microcomputer is designed for the characteristics of expensive, complex functions and inconvenient portability of the phase meters on the market. The phase meter is mainly composed of STM32 single-chip microcomputer module, amplification and shaping module, display module, power supply module and so on. The article first introduces the design background and goals of the phase meter; secondly, introduces the overall design scheme and the selection of each module; thirdly, makes a real object for frequency and phase function tests; finally, makes a design summary. Experiments show that the frequency and phase measurement errors of the low-frequency phase meter are small, and the expected design goal is achieved. At the same time, the phase meter can measure a variety of waveforms, with low production costs and easy portability.

With the rapid development of high-voltage transmission engineering, more and more cranes are operated around the high-voltage lines. Due to the high-altitude construction crane is too close to the high-voltage transmission lines, some accidents of the transmission lines often occur such as the short circuit and tripping, which has brought great impact on daily life and production. Therefore, it is necessary to design a system about anti-collision alarming against high-voltage transmission lines to prevent damages to operators. Considering the above situation, a system about anti-collision alarm of crane against high-voltage lines based on multi-modal information is proposed in this paper. The framework of this system is mainly divided into perception layer, detection layer and application layer. Through the simulation analysis and experimental verification of multimodal information, it indicates that our proposed system has a good effect both in real detection and anti-collision, which can effectively avoid the occurrence of accidents in the process of crane operation.

Online English teaching app is a platform that can make students more flexible and effective in mastering English skills. At the same time, teachers can be more targeted and motivated in the classroom, making the guidance more targeted. This paper sorts out the basic theory of teaching quality evaluation and introduces the advantages and disadvantages of the more commonly used evaluation methods such as fuzzy comprehensive evaluation, AHP and BP neural network evaluation. In this paper, the relevant information of three random teachers in the English online teaching platform was obtained by means of a questionnaire survey, and then the teaching quality of the three teachers on the English Teaching Platform (EOTP) was simulated and analyzed in MATLAB, and the simulation results were compared with the teaching quality of the three teachers. The results of fuzzy comprehensive evaluation method and BP neural network evaluation method are compared and analyzed. The analysis results show that the teaching evaluation model based on grey relational analysis has certain rationality and feasibility.

Causality is the relation of inevitability between violations of service and failure-factors. It is evidence of accountability in cloud service because compared with correlation, the causality shows more probative capability in evidentiary capability. This paper surveys the causality in recent research, including the features, patterns, and hypotheses of causality, and also the methods of discovery and analysis. Most research focuses on a single causal feature and the patterns of causality in typical violations, such as long tails, lack of resources, etc., and need to explore further the differences in causal patterns in different situations. At last, the results are proposed in this paper, including a taxonomy and some challenges in recent research, and an approach to causality analysis.

This paper extracts the maintenance records of the equipment in the equipment big data platform, uses the FP-growth algorithm to mine association rules, obtains typical association rules, and mines the associations of various faults occurring in regions, motorcycle hours, etc., to provide decision support for maintenance plan formulation, maintenance equipment procurement and storage, and to provide technical support for improving the effectiveness of equipment support.

For industrial CT, the existing 3D measurement method based on industrial CT sequence imaging needs to be improved in terms of accuracy, speed and the degree of automation. In this context, based on the existing imaging measurement theory and taking the industrial CT sequence imaging as the experimental object, two volume measurement methods have been proposed this paper based on the industrial CT sequence imaging. This method was applied to carry out equal spacing stratified CT scan on three typical shapes of standard steel workpiece and Actual workpiece with inner cavity, and the obtained sequence fault imaging was measured and analyzed and verified. The experimental results show that the two measurement methods proposed in this paper improve the accuracy of measurement by two to three orders of magnitude and the accuracy of measurement is better than 1% compared with the existing volume measurement algorithm, which can meet the high precision requirements of the engineering application.

The maximum current speed distribution of ocean surface currents during the period of severe tropical storm Lionrock is investigated using six days’ currents data measured by High Frequency Surface Wave Radar (HFSWR) in the southwestern Taiwan Straits. Analysis shows that an increment up to about 100-cm/s is induced by the Lionrock at the southern and northeastern area, and the increment is nearly symmetrical about the track of Lionrock. The increments of the maximum currents are larger at the right of the track of Lionrock than that of the left.

The technology for recognizing and detecting man’s abnormal behavior in motion is a hot issue in the field of computer vision. The analysis of human movement characteristics is a crucial component of human body detection and has garnered significant attention from scholars. This process involves the use of technical means to obtain motion parameters of the human body, enabling the reconstruction of its structure and posture, and ultimately facilitating the comprehension and application of human movement. Currently, the video surveillance system is mainly designed to detect and track moving targets in the scene, while the function of identifying and recognizing the targets is poor. Under the circumstances, when emergencies occur, the video surveillance system can hardly give a real-time report on the emergencies. Thus, it is really important to design a video surveillance system that provides real-time detection and recognition of abnormal behavior of moving targets. To this end, this paper gives a review of various methods of detecting abnormal behavior based on man’s motion features. Firstly, it summarizes different methods of detecting human bodies in motion. Secondly, it sums up techniques of extracting motion features under the condition of analyzing motion features and analyzes the motion feature. Thirdly, it discusses and analyzes common algorithms of abnormal behavior detection and gives a dataset in the field of motion feature analysis. Finally, it depicts the outlook for the future study of the research field the paper discusses.

RGB-D object detection is a challenging task due to the demand of effectively processing of visible modality and depth modality features. However, pre-existing RGB-D object detection models have several deficiencies, including demand for hand-crafted settings, and insufficient ability of fusing cross-modal features. In this paper, we propose a novel Cross-modal RGB-D object detection model, based on Deformable DETR, named as CM-DETR. Our proposed model can effectively fuse multi-modal information, and don’t need hand-crafted settings resulted from prior information. Extensive experiments show that our model has achieved extraordinary improvement, which exceeds the baseline by more than 4.6% mAP on SUN-RGBD and 6.9% mAP on NYUDv2.

In recent years, the impact of air quality on people has gradually increased, and how to effectively analyze the changing laws of air pollutant components is of great significance. Due to the problems of vanishing gradients and memory constraints, most existing methods for long sequence prediction struggle to exploit the full contextual information of the sequence. This paper introduced an improved Autoformer model based on the autocorrelation mechanism, which divides the sequence into a periodic sequence and a cyclic trend sequence through time series decomposition, and further extracts deeper hidden features from the data by adding a feature extraction module to the architecture. Use the trend module to process the cyclic trend sequence to increase the model's perception of the trend. The experiments prove that the model has good prediction effect.

In this paper, the radar servo system is modeled based on PSO Algorithm and Akaike Information Criterion and system identification theory. It is modeled by linear dynamic equations. Linear model of parameters which have a variety of different orders are identified based on PSO Algorithm whose order can be decided by Akaike Information Criterion. At last, the linear mathematical model of radar servo system can be concluded. The test and the simulation results show that it has great effect that radar servo system is described by this model.

In order to understand the application effect of zero trust security monitoring of intelligent distribution network terminals, research on zero trust security monitoring of intelligent distribution network terminals based on 5G communication technology was proposed. This paper proposes a design scheme of distribution network automation terminal based on embedded system and with multiple communication modes. The terminal collects the effective value of three-phase voltage and current, three-phase voltage, three-phase active power, reactive power, apparent power, frequency, and other power parameters. It can also collect eight analog signals at the same time. It has five communication modes: RS-232, RS-485, GPRS, CAN, and Ethernet. Users can choose any communication mode according to the actual situation, thus ensuring the compatibility of various communication interfaces. In the context of smart grid and energy internet construction, the construction of smart grid cannot be separated from communication technology, which has a direct impact on the intelligent level of power equipment. The application of 5G communication technology in smart grid can promote the better and faster development of smart grid.

In the process of text detection, we frequently encounter numerous indistinct images, which can easily result in text omission and misdetection. Inspired by the SPSR model, we introduce gradient branching to guide the training of text detection models in order to address this problem. By preserving more image edge features, we expect to improve the text detection performance of fuzzy images and fuzzy regions. The experiment demonstrates that the gradient guidance-based text detection model can detect text in ambiguous images more accurately and reduce instances of missing and incorrect detection.

Directed greybox fuzzing aims to test specific code and has made many advances in several areas. However, most vulnerabilities of input parsing programs are triggered in the particular state of the program, so existing directed greybox fuzzing works face path explosion problem when they fuzz the input parsing program and need more ability to explore the particular state of the program. To address the above problem, we propose a call-relationship-based fitness function. The main idea is to use the function call relationship to guide directed fuzzing before reaching the target. Call-relationship-based fitness function extracts the function calls and call relationship from the program, uses an intra-procedural reachability analysis to get all concerned edges, and constructs the fitness function based on the edges. Based on the above method, we implemented the directed greybox fuzzing IPDF and evaluated it with the mainstream directed greybox fuzzers Beacon and AFLGo on real-world programs. Evaluation of IPDF showed that IPDF found vulnerabilities faster than the state-of-the-art directed greybox fuzzers. The experimental results showed that the speed of MDGF is 3.01 times faster than that of AFLGo and 1.15 times faster than Beacon.

Pipeline plays an important role in petrochemical production, and its health monitoring is related to safety production. At present, the petrochemical production mainly relies on manual inspection to judge the pipeline status, which cannot achieve the full-time domain health monitoring. The purpose of this paper is to propose a method that uses a single weak fiber grating array as a sensor, extracts the vibration signal generated in the pipeline operation by the phase-sensitive optical time-domain reflection system, and improves the LSTM algorithm by using the variational mode decomposition principle and the arrangement entropy theory to predict the pipeline operation signal, so as to realize the abnormal state early warning. The experimental results show that the improved VMD-PE-LSTM model proposed in this paper has a very high accuracy in the prediction of pipeline vibration signals and provides a new direction for the early warning of abnormal pipeline operation.

With the proposal of dual-carbon economy, smart grid is developing towards energy conservation and emission reduction, and the abnormal power consumption of users causes serious loss of power resources. Aiming at the problems of low accuracy and slow operation efficiency of traditional abnormal power consumption detection methods, a new method combining Light GBM model with improved short-term and short-term memory network model is proposed for abnormal power consumption detection. Anomaly detection is carried out by combining sampling and Lightgbm model, and the abnormal power consumption category is given by improving the short-term and short-term memory network model. The advantages of the proposed method are analyzed through experiments. The results show that compared with traditional detection methods, this method can detect abnormal users quickly and effectively, and the detection accuracy is 97.83%. At the same time, the abnormal data is effectively classified, and the comprehensive classification accuracy is 96.61%. It provides a certain reference for the development of anomaly detection technology.

To reduce transformer faults to ensure the safe operation of the power system, and at the same time avoid external interference to improve the accuracy of monitoring, this paper proposes a wireless detection method for dielectric loss of substation capacitive equipment based on space laser communication technology. Firstly, the data samples of space laser communication network information are collected, the dielectric loss factor is calculated by using the harmonic analysis method, and the data samples are pre-processed, then artificial intelligence is used to design the space laser communication dielectric loss detection model, and finally, the space laser communication dielectric loss detection experiment is carried out. The detection experiment results show that this method can effectively improve the correct detection rate of the dielectric loss factor.

Point cloud registration is an important part of point cloud processing. An unsuitable down sampling filter in registration can lead to inaccurate registration results. In this paper, a Curvature-Based Segmental Sampling (CBSS) method is proposed, which arranges the curvatures of vertices in descending order and then divides them into many segments, and finally selects the vertices with higher curvature as the sampling results. The algorithm is implemented with the open-source library Point Cloud Library (PCL) and then the filter is compared with Voxel Grid (VG) and Normal Space Sampling (NSS) through a series of registration experiments. The experimental results demonstrate that the proposed filter outperforms the other two filters in achieving higher coarse registration accuracy.

In this paper, a transmission scheme of rate-split multiple access system in a power line communication channel is considered. The scheme consists of one sender and two users, and the signal for all users is superimposed at the sender. Using the continuous convex approximation method, we aim to maximize the total achievable rate of the system by optimizing the beamforming vector and rate allocation, and then propose a method based on continuous convex approximation to solve the optimization problem. The results show that, compared with PLC-NOMA, PLC-RSMA is suitable for various channel directions (whether the channel directions are aligned or orthogonal). PLC-RSMA can achieve better transmission rate performance, and with the increase of SNR, the rate performance gain of PLC-RSMA becomes more and more obvious. At the set parameters, when SNR is 30dB, the relative rate performance of PLC-RSMA is improved by at least 37% compared to PLC-NOMA.

Aiming at the characteristics of complex biochemical reaction, nonlinearity and difficult prediction of dissolved oxygen in sewage treatment process, this paper proposes a dissolved oxygen concentration prediction model based on CNN-BiLSTM hybrid artificial neural network. Firstly, the abnormal data is identified and eliminated by data preprocessing, and the missing data is filled by interpolation method. Then, the Pearson correlation coefficient is used to analyze the correlation between dissolved oxygen and other variables. Multiple variable data with good correlation are selected and input into the CNN-BiLSTM network model. The dissolved oxygen concentration is predicted by CNN convolution operation combined with bidirectional long-term and short-term memory neural network (Bi-LSTM), and the time attention mechanism is introduced to learn the weight distribution between different time steps, focusing on the time step that has the greatest impact on dissolved oxygen concentration, so as to improve the prediction accuracy of the model. Compared with LSTM, GRU, CNN-LSTM and CNN-GRU models, the simulation results show that the proposed model can predict the dissolved oxygen more accurately and has higher prediction accuracy.

The DC resistivity method uses the inherent conductive differences of subsurface materials as the basis for exploration to observe the earth's electric field, while human activities, nature and solar storms can cause changes in the earth's electric field, which in turn seriously affects the accuracy of exploration. Today the DC resistivity method is used for monitoring hazards such as landfill seepage, coal mine seepage, and slope seepage. The above hazards change slowly, and the monitored change signal is weak, and if there is earth current interference around it will bury the hazard signal. Therefore, it makes sense to noise cancel the signal for DC method. In this paper, Empirical Mode Decomposition (EMD), sliding average and least squares fitting are combined to propose a noise-canceling method for DC electrical law signals. In this paper, the effectiveness of the method is verified by theoretical simulations.

With the increase of voltage level and capacity of modern power system, three-level converter, as the main circuit of active power filter, meets the requirement of improving compensation capacity. However, the number of power devices of three-level converter is twice as large as that of two-level converter, and the probability of power device failure is greater, which affects the normal operation of the system. Aiming at the fault of three-level active power filter, this paper studies the fault characteristics of different power components of the main circuit T-type three-level converter when open-circuit fault occurs, applies the hybrid theory to the fault diagnosis of the converter, establishes the normal model and the fault model of different power components when open-circuit fault occurs, and realizes the location of the fault components according to the residual generated by the predicted output and the actual output of the model, the effectiveness of this method is verified by simulation.

In this paper, a virtual power plant model considering demand side response is proposed. Considering the influence of price demand-side response on virtual power plant, an optimal scheduling control method was proposed for virtual power plant, which included demand-side response, wind power, photovoltaic, gas turbine and energy storage equipment. Firstly, a mathematical model was built with the maximum net income of virtual power plant as the objective function. Secondly, a reasonable operation strategy was designed. The demand-side response strategy was optimized based on wind-landscape output prediction. Thirdly, the gas turbine and energy storage equipment coordinated the output allocation to compensate the deviation between the actual output and the predicted output of the scenery. Finally, through the simulation analysis of the example, the results show that the consideration of the demand-side response based on price in the virtual power plant can standardize the user's electricity consumption. And the load peaking and valley filling effect is obvious. The designed operation strategy has coordinated the power distribution of various energy sources and maximized the income of the virtual power plant.

As an important direction of energy reform, comprehensive energy system has become the focus of research and attention in the world. Improving the efficiency of comprehensive energy utilization and bringing good social and economic benefits as well as social and environmental benefits is the only way for the country to achieve high-quality economic development, and it is also the current national policy. But what are the ways to use energy in a comprehensive way? So, in addition to improving the energy utilization rate of equipment, reduce the energy loss of each production link, but also need to take corresponding management measures and technology to reduce energy consumption. Based on this, the acquisition and detection of comprehensive energy can effectively promote the development of energy conservation, emission reduction and low-carbon economy. In this paper, for the comprehensive utilization of energy, a feature detector is introduced into the comprehensive energy system to monitor the comprehensive energy processing time cycle, power consumption and other performance parameters. Experimental results show that: under the condition of appropriate sensing threshold, the feature detector has a good time sensing effect, which is efficient and feasible for energy collection and monitoring. It also provides a theoretical basis for the supervision and use of the production energy management department, so that the management department can grasp the energy consumption situation of the enterprise at any time and make timely scheduling adjustment according to this.

The number of channels for radio frequency simulation system is very limited, and the traditional simulation method of one channel simulating one target is difficult to meet the demand of complex scene simulation. This paper proposes a simulation method based on the triadic vector control algorithm of "both amplitude and phase control" to simulate the angular position of multiple radar targets by one channel, and further researches the simulation method based on digital radio frequency memory technology to simulate the multidimensional characteristics of multiple radar targets such as delay, doppler frequency, amplitude and angle by one channel. Mathematical simulation experiments are designed to verify the feasibility of the method.

In recent years, as a clean and efficient energy storage technology, lithium iron phosphate battery is widely used in large energy storage power stations, new energy vehicles and other fields. However, lithium-ion batteries still face obstacles that limit their application space. Once the temperature exceeds the working range of the battery, lithium iron phosphate battery will rapidly degrade, increasing the risk of facing fire and explosion and other safety problems. Multi-physical field full-coupling simulation is one of the important means to study the temperature of lithium iron phosphate battery operation process. Based on the electrochemical-thermal coupling full model of iron phosphate battery, this paper analyzes and studies the temperature change of battery operation process and obtains the temperature change of battery under adiabatic and actual conditions, which provides a basis for solving the thermal management problem of lithium iron phosphate battery operation process.

To solve the reconnaissance problem of Linear Frequency Modulation (LFM) pulse signal, this paper proposes an estimation method for the full set of parameters of frequency modulation slope, initial frequency, pulse repetition period, duty cycle, and pulse width. Firstly, the Fractional Fourier Transform (FRFT) is used to estimate the frequency modulation slope and the initial frequency of the signal. Secondly, the method of periodic accumulation is used to improve the signal-to-noise ratio, and a specific threshold value is set to estimate the duty cycle. Finally, the parameters of pulse repetition period and pulse width are estimated by autocorrelation function method. The simulation results show that the method can accurately detect the relevant parameters of LFM pulse signal.

Oceanic eddy is a highly occurring marine phenomenon in the ocean, characterized by closed circulation. Oceanic eddy can cause changes in Sea Surface Height (SSH) and will make the SSH around the center of the oceanic eddy to be higher or lower than the surrounding area, so the identification of oceanic eddy can be achieved by detecting the change of SSH. Traditionally, radar altimeter is an important means to measure the sea surface height, but its resolution is low, so it can’t effectively detect sub-mesoscale oceanic eddies. With its all-time, all-weather, wide-swath and high-resolution characteristics, Synthetic Aperture Radar (SAR) has become a research hotspot in oceanic eddy remote sensing detection. It uses radar antennas from different angles of view to form a baseline, called interferometric SAR, which can achieve the measurement of SSH. In this paper, a signal level simulation model of oceanic eddy for spaceborne cross-track interferometric SAR(XTI-SAR) is established, and an evaluation method based on ideal interferometric factors is introduced. In the experiments, we analyze the influence on the error of oceanic eddy SSH measurement at different sea states. The results show that the increase of wind speed will increase the height measurement error of oceanic eddy, and the difference of wind direction will cause obvious changes in the height measurement error.

The article establishes a two-layer optimization model, in which the user-side load and the microgrid operation are combined. The upper layer of optimization establishes a demand response model based on the adjustable characteristics of the load and the time characteristics of the time-sharing tariff. To give full play to the characteristics of the demand response that can cut the peaks and fill the valleys, to reduce the energy cost on the user side and improve the energy utilization, and the lower-level optimization is based on the best demand response results obtained from the upper layer of optimization. Based on the supply-demand balance, Regulating the output power of distributed energy sources at various times of the day, in order to reduce the cost of microgrid system operation. Finally, the proposed method is validated by solving the two-layer optimization model with an improved particle swarm algorithm.

Reverse Time Migration (RTM) is one of the main seismic migration imaging algorithms, which has the advantages of high imaging accuracy and being unrestricted by tilt angle. In this paper, RTM based on cross-correlation imaging condition is used for electromagnetic waves imaging. After reducing low frequency noise using wavefield decomposition, the results of electromagnetic waves imaging still contain noise caused by multipath effects and clutter from the background. In order to reduce the effects of noises, we propose an improved RTM imaging method based on singular value decomposition (SVD). The numerical results show that the proposed method can reducing multipath noise and background clutter effectively by RTM combing wavefield decomposition with SVD. Electromagnetic wave imaging quality is improved.

Rat hole detection is a key work in the prevention of rat damage. The digital rat hole detection method of UAV combined with target detection is studied to replace the manual carpet rat hole detection, so as to improve the detection efficiency. In this paper, low-altitude remote sensing of unmanned aerial vehicle was used to collect rat hole images on the Edolechuan grassland in Hohhot, Inner Mongolia. Combined with deep learning models: Faster-Rcnn, Yolov3 and SSD, rat hole detection was compared and analyzed. Data was cut and labeled through image preprocessing method. Then, by comparing the three groups of target detection models, the results show that the SSD model has the best effect on rat hole detection, the accuracy rate can reach 91.8%, and the reasoning speed can reach 7.9ms.

During the outbreak of COVID-19, many cities adopted strict travel measures to quickly cut off the transmission chain. However, these measures hindered the development of the urban economy as time passed. In the face of the possible recurrence of the epidemic, how to develop reasonable operational measures to ensure commute trips and guarantee the normal operation of the urban economy has become an important issue facing the current transportation system. Therefore, this paper is based on the smart card data to explore the spatial-temporal characteristic changes of residents' commuting under the influence of the epidemic. Firstly, the temporal characteristics of each epidemic period are described from three dimensions. Then, travel hotspots are identified and the spatial distribution changes of them in each epidemic period are explored. Next, the correlation coefficient between hotspot metro stations and surrounding land use is calculated. Finally, the rules are summarized, and operational suggestions are proposed. This article analyzes the commuting characteristics of residents based on smart card data, which can provide data support for optimizing metro operation measures.

Aiming at the problem that multi-signal time-frequency domain overlapping is very likely to appear in the actual scene, it is difficult to identify, and a broadband signal intelligent detection method based on time-frequency analysis and target detection network is proposed. The communication signal is transformed into a time-frequency image, and the YOLOv5 network model is improved. Introduce the attention mechanism in the feature extraction network, highlight the signal area information, modify the K-Means clustering rules, recalculate the size of the prior frame, and use the improved model to study the time-frequency diagram of multiple signals overlapping. The results show that under different signal-to-noise ratios, the detection and parameter estimation of six types of random overlapping communication signals are realized. Overall, the signal detection probability reaches 94.74%, the false alarm probability is 2.86%, and the average error of parameter estimation is 2.15%. The analysis results in this paper can provide support for the rapid and effective detection of signals in practical applications.

Laser ultrasonic backscattering is of great significance to nondestructive evaluation of the microstructure of strong-scattering metal materials. However, it is difficult to extract the microstructural backscattering signals because of high-level electrical noises. In this manuscript, the laser ultrasonic backscattering characteristics of metal microstructure are analyzed and studied based on the Empirical Mode Decomposition (EMD). TA2 titanium alloy was heat-treated at 800°C for different times to obtain single-phase titanium alloy specimens with different grain sizes. The laser-ultrasonic waveforms transmitted by the bottom surface of the sample were obtained. The EMD was performed on the backscattering noise signal between two successive pulse echo signals, and the multi-order Intrinsic Mode Function (IMF) was obtained. The correlation between the average spectrum of multiple measurements of each order IMF and the spectrum of the first longitudinal wave pulse echo was analyzed. and the IMF with the largest correlation coefficient was selected as the effective IMF, which was the most relevant to the microstructure. Subsequently, the backscattering levels of each sample were calculated based on the variance analysis of the effective IMF measured at multiple points. The results have shown that the backscattering level is positively correlated with the grain size of the metal structure in a specific frequency range.

Digital watermarking is a technique to embed specific digital signals into digital products to protect copyright integrity, which is an effective way to protect information security. Neural network technology is a mathematical model generation technology that mimics the structural function of biological neural networks and has good self-learning and merit-seeking capabilities. It is feasible to add neural network to digital watermarking technology for copyright protection research. In this paper, we select and improve a generative adversarial network from a classical neural network to generate adversarial samples required for subsequent research using watermarked datasets, where the improvements include combining high-level loss and low-level feature loss and using adversarial labels to facilitate adversarial classification. The improved generated adversarial samples have the advantage of less alteration to the original image. In this paper, new digital mosaic techniques combined with neural networks are explored in depth rain for new big data directions, which are informative for new areas of research.

The factors affecting the transmission performance of LORA technology are further studied to deepen the analysis and understanding of the quality of the communication links of LOAR technology. In this paper, we study the transmission experiments of the LoRa technique at different emission powers and at different aerial speeds to verify its communication quality. The experimental results show that the communication distance and communication quality, which is the signal packet loss rate, are significantly changed with the emission power and air velocity. Then, the power consumption of LORA transceiver is studied and tested in this paper, which shows a linear relationship between node power consumption and air rate.

To solve the problem that radar can still generate effective tracks to the covered target under the existing false target jamming technology, a method of jamming radar target track generation using finite false target signals is proposed through the plot and track correlation process of the radar in the initial stage of target searching and tracking. Through electronic reconnaissance, interception, and retransmission of radar signals by a jammer, a finite number of false target signals with random movement characteristics are generated around the covered target, interfering with track correlation during target data processing by radar. Simulation examples demonstrate the effectiveness of the proposed method.

The automatic commodity classification system as an important part of e-commerce platform is getting more and more attention. In this paper, we take electronic commodity text title as the research object, deeply analyze its text characteristics, and propose the commodity classification model Bert-VAE with enhanced commodity text title features. Firstly, to address the problem that the current short text classification methods cannot take into account both global and local information on the acquisition of information by word vectors, the Bert model is used to obtain text feature representations. Then, the text feature enhancement method based on Variational Auto-Encoder (VAE) is proposed to address the problems of poor discriminative text titles and unbalanced sample sizes of various categories of commodities. Finally, deep text features and enhanced features are fused to predict commodity categories. In order to verify the model performance, experiments are conducted on the e-tailing commodity dataset, and the experimental results show that the commodity classification model proposed in this paper outperforms other current models in this task to a certain extent.

Most of the current existing style transfer methods are based on photographs or Western paintings. Due to the inherent differences between Chinese and Western paintings, direct application of existing algorithms cannot generate satisfactory style transfer results for Chinese ink paintings. Accordingly, this paper proposes a new feedforward style transfer method, called inkStyle, which uses a draw Network to transfer global style patterns at low resolution and a higher resolution details networks to modify local style patterns in a pyramidal manner based on the multi-level Laplacian filtering output of the content image. The experimental results show that the method in the paper performs better and produces better visual results.

An improved timing synchronization algorithm based on ZC (Zadoff-Chu) sequence is proposed in order to improve timing synchronization accuracy of OFDM (Orthogonal Frequency Division Multiplexing) system. This algorithm combined the sequence inversion of Minn algorithm with the sequence conjugate of Park algorithm. The new sequences replaced PN sequences with ZC sequences that have good autocorrelation and low cross-correlation. Even in the timing position error, the system can still be precisely timing synchronized. Under the channel of AWGN (Additive Gaussian White Noise), the number of subcarriers is 25616QAM modulation and demodulation. Matlab simulation shows that the improved algorithm eliminates the influence of minor peak on timing synchronization in Minn algorithm and Park algorithm Timing deviation and MSE (Mean Square Error) are lower than Minn algorithm and Park algorithm within -15 to approximately15-dB, which improves timing accuracy. When the SNR is 15-dB, the timing MSE of the improved algorithm is 4.2×10^-6, which is the optimal state for estimating performance.

Aiming at the problem that the accuracy of abnormal sound detection under unsupervised conditions is not ideal, a novel abnormal sound detection model using composite self-coder combined with Gaussian mixture model is proposed. Firstly, the timing structure and gating mechanism of LSTM are used to improve the feature extraction ability of self-coder (including self-coder and variational self-coder), Secondly, Gaussian Mixture Model (GMM) is used to generate artificial data to improve the robustness of the self-coder against background noise. Experiments are carried out using ToyADMOS and MIMII public data sets, and the results are superior to the naive self-coder and the two improved self-coding models. On the six machines of the experimental data set, AUC increases by 6.34%, 6.65%, 4.03%, 5.57%, 2.38% and 1.07% respectively.

With the rapid development of the economy and the continuous improvement of people's living standards, the production of daily waste has increased sharply. In order to effectively realize waste reduction and resource recycling, garbage classification has been widely promoted. However, despite various publicity efforts, there are still problems with unclear, ambiguous, and incorrect garbage classification among residents. To address these issues, an intelligent garbage classification system based on vision and speech recognition technology has been designed. The system includes an intelligent management terminal at each park's garbage collection point, which collects voice and image information for processing and analysis on a central server. The terminal is equipped with the ability to control the opening and closing of the intelligent garbage can lid and monitor the status of the garbage collection points. Furthermore, a supporting mobile application is available, enabling users to learn about garbage classification knowledge online and to carry out proper garbage classification. The implementation of this intelligent system solves the problem of garbage classification, which improves the efficiency of garbage treatment and provides significant practical and social value.

In the vector operation system, each data operation will only increase or decrease the same number for each component of the vector, that is to say, the vector is regarded as a data object on the whole, and each operation is carried out on the whole. For example, translation, scaling and other operations in computer graphics. Such computing systems are currently widely used in graphics, machine learning, mathematical modeling and other fields. When the data in the vector computing system is homomorphic encrypted and uploaded to the cloud server, the problem of low vector coding efficiency and polynomial coefficient expansion caused by a large number of polynomial operations is generated. The existing SIMD coding algorithm encodes an n-dimensional vector integer into a polynomial. The number of calculations increases exponentially with the increase of n, and the coding efficiency of the vector is very low. For typical FHE schemes such as BGV, BFV and CKKS, the main performance bottleneck comes from a large number of polynomial algorithms. Specifically, encrypted data is typically composed of a pair of polynomials with coefficients of hundreds or thousands of bits, requiring expensive multiword arithmetic. In addition, large polynomial lengths increase the computational complexity. Therefore, in this paper, a vector compression coding algorithm based on SIMD(VCABSIMD) is proposed to solve the performance problem when the data in the vector computing system is homomorphically encrypted and uploaded to the cloud server for calculation. The vector coding efficiency and polynomial calculation are deeply optimized.

Aiming at the low precision of the traditional binocular stereo matching algorithm in calculating the matching cost of the strong texture area, a binocular stereo matching algorithm based on a dynamic tilted window is proposed. First, the absolute value of brightness or color difference is replaced by random initialization of pixels, and then the traditional Census cross-domain transformation is replaced by a dynamically tilted disparity plane. For the traditional algorithm, the matching accuracy is improved, making the window more adaptable to the actual environment. In the cost calculation step, a gray histogram is added as an indicator for judging the texture difference, which improves the matching cost in the strong texture area; on this basis, iterates from space propagation, plane propagation to view propagation. In the parallax optimization step, left-right consistency detection and parallax filling are used to further optimize the reduction of the false matching rate. The experimental data is compared with the standard images on the Middlebury dataset. The results show that the average error matching rate of the disparity map generated by the stereo matching algorithm of the dynamic tilted window of this method reaches 4.03%. Compared with the Census algorithm, the matching error rate is respectively reduced. 21.1%, effectively improving the matching accuracy; compared with other algorithms, the false matching rate for high texture areas increased by 1.2% and 3.71%.

At present, text sentiment multi-classification model has problems of insufficient semantic feature fusion and ignore the syntactic structure of sentences. Therefore, this paper proposes a dual graph convolutional network model, which extract semantic and syntactic information of text by semantic graph convolutional network and syntactic graph convolutional network. Also, this paper proposes a label graph embedding method to fuse richer semantic features. Finally, experiments on two public datasets show that our method achieves better results.

In modern chip industry production, the existing object detection algorithm parameters are large in number and complex in structure, which is gradually failed to meet the high standards in the production of contemporary enterprises, for the above testing problem, the defect detection algorithm of lightweight chip based on improved YOLOv5s was proposed. First, make improvements on the anchors clustering algorithm to change the definition of the distance between the sample data, and the feature extraction mechanism of ShuffleNetv2 is introduced in the backbone network, reduced the calculation of parameters, at last, the loss function of Focal-EIOU is used, which enhanced the detection performance of YOLOv5s network for various chip defects. The mAP_0.5 of the modified network can reach 87.6% on the provided chip dataset, compared with other network structures there are obvious improvements, and the number of network parameters decreased by 39.7%, which fully proves that the improved network can reach the high standard and lightweight in chip defect detection.

According to the characteristics of space camera subject to working environment and task requirements, the automatic exposure and automatic white balance are analyzed theoretically. Gaussian weight is used to correlate the target position and gray value of image, and the relationship between variable step size adjustment, balance accuracy and speed of automatic exposure is optimized. Automatic white balance performs color temperature estimation in RGB space, enabling accurate image color restoration under different illumination. The optimized design of hardware implementation according to the FPGA characteristics. Through the simulation software analysis and the effect of the hardware system display pictures, it is shown that the FPGA hardware system can run high-definition real-time, fast speed, low power consumption and strong practicability. This hardware system has been used in image processing of space cameras.

In order to strengthen the management of experimental teaching process and improve the accuracy of abnormal behavior detection of students, the existing YOLOv4 algorithm is improved. By adding channel attention mechanism SE module in the main feature extraction network, the algorithm ensures fast and effective extraction of student behavior effective features. A soft pooled SPP network based on CSP structure is used to assist optimization and feature extraction to reduce information loss in the pooling process. By introducing Focal loss function to deal with uneven number of plus and minus samples and difficult sample training in data, the effect of student behavior classification is improved. The algorithm is verified by experiment on the self-made behavior data set. The results show that the improved YOLOv4 algorithm has a better detection effect on students' behavior, and the average accuracy rate reaches 86.41%. Compared with the YOLOv4 algorithm, the recognition accuracy rate is improved by about 7%.

We proposed a pedestrian detection algorithm combining YOLOv5 with convolution and channel attention mechanism. First, we use our own pedestrian dataset to train the YOLOv5 detection model. Then, three attention mechanisms, SE, CBAMC3, and CoordAtt, are used to enhance the detection performance. The experiment demonstrated that the precision of both SE and CoordAtt decreased, the recall of SE also decreased, while the accuracy of CBAMC3 was improved and the mAP changed little, thus CBAMC3 became the best model for pedestrian detection. Research indicates that adding convolution block attention modules can increase the precision of detecting small pedestrian targets.

In the real world, the resolution of the image that is collected can vary depending on the camera's quality or the change in the distance from the pedestrian. Important information is lost from the low-resolution image. It can be difficult to match Low Resolution (LR) input photographs with High Resolution (HR) gallery images. Thus, we suggest that the super-resolution module and the multi-feature extraction module be improved in order to address the aforementioned issues. To be more precise, the resolution of the low-resolution query image is restored in the first step using an upgraded Super Resolution (SR) model (VDSR-NAM). A two-stream feature extraction network extracts and fuses the features of the LR and SR images in the second stage. The potential of our model has been shown in numerous tests on cross-resolution person re-id datasets. The efficacy of the loss function on our model is concurrently confirmed by ablation experiments on the dataset MLR-VIPER.

In this paper, we propose a foggy weather object detection model based on an attention mechanism, to address the problem of low detection accuracy, missed detection and false detection when general object detection models are applied directly to foggy scenes. Firstly, to enhance the detection network's multi-scale expression ability and sensitivity to the target, a residual module that integrates the attention mechanism replaces the BottleNeck module of the backbone network. This design improves the network's ability to extract features and locate targets at a fine-grained level. Secondly, the CIOU loss function replaces the original loss function, improving the stability of the bounding box regression process. Thirdly, the K-means++ clustering algorithm is used to generate anchors suitable for the dataset in this paper. Furthermore, the object detection dataset in foggy scenes is further enriched based on the atmospheric scattering model. Experimental results indicate that the proposed method's mAP in light fog, medium fog and dense fog scenes is increased by 7.4%, 6.05% and 6.36%, respectively, compared to the original YOLOv5s. This improvement in accuracy significantly reduces the missed detection rate and false detection rate, effectively enhancing object detection performance in foggy weather.

The clustering ensemble is formed by combining clustering analysis and ensemble learning. However, most clustering ensemble methods treat all samples equally, which negatively affects the final clustering result. To this end, we propose sample’s representation to evaluate the sample’s importance in clustering comprehensively. The sample’s representation measures the clustering importance of samples from two perspectives: the stability of the relationship between the sample and its neighbor samples and the closeness of the relationship between the sample and its neighbor samples. According to the representation of each sample, we divide a dataset into cluster core and cluster halo. Then we obtain the credible underlying structure through the cluster core samples. Finally, the cluster halo samples are gradually allocated to the above structure to get the final clustering result. The working steps of the algorithm are shown on two synthetic datasets, and experiments on nine real datasets fully demonstrate that the algorithm outperforms 11 other state-of-the-art clustering ensemble methods.

Handwriting Chinese Character Recognition (HCCR) is the foundation of document digitization. It is a challenging subject in the field of image classification and recognition for a series of reasons such as the large number of Chinese characters, the diversification of writing style and numerous similar characters. To solve the above problems, this paper designs a four-channel convolution recognition model based on MobileNetV2. First, the input image is sent to four-channel convolution with different receptive fields, and feature maps of different scales are extracted respectively to improve the accuracy of the model. Then the feature maps are combined to enrich the diversity of features. Afterwards, the combined features are weighted by SE Block, and more useful feature maps are screened by this means to accelerate the model convergence. Finally, the lightweight network mobilenetv2 is used to classify the weighted features. The experimental results show that the recognition accuracy of the four-channel convolution recognition model based on mobilenetv2 on the offline handwritten Chinese character set CASIA-HWDB1.1 has reached 96.05%, and the convergence speed of the model is extremely fast. Also, the memory occupation and parameter quantity are far lower than other Chinese handwriting character recognition models.

Image segmentation is a classical and basic problem in the domain of computer vision. Due to the fact that fully supervision segmentation methods require dense time-consuming and expensive manual-annotations, lots of WSSS (Weakly-Supervised Semantic Segmentation) methods have been proposed to take advantage of the simplicity and availability of weak supervision annotations. In this work, we build an integrated framework to jointly train the classification task and the segmentation task guided by a self-supervised thinking with only image-level supervision and a compound refinement strategy. Then, we introduce a restrictive adversarial erasing approach to push our model to find more segmentation cues. We evaluate the proposed method on PASCAL VOC 2012 benchmark, and the experiments show that our method can achieve competitive performance compared with the earlier methods.

In order to reduce the total identification time of readers and improve the system throughput in Radio Frequency Identification (RFID) systems, a new algorithm is proposed. Based on the Frame-Slotted ALOHA (FSA) algorithm, Tags are grouped according to the number of time slots during the reader's recognition process. For tags in collision time slots after a frame ends, the Collision Tree (CT) algorithm is used for group recognition, solving the starvation problem of the FSA algorithm and significantly reducing the total identification time of the reader. Simulation experiments have demonstrated that the algorithm proposed by the author significantly reduces the processing time when compared to the CT algorithm. and achieves a throughput rate that is higher than the traditional CT algorithm When the number of tags exceeds 260, reaching a maximum of 52% when the number of tags is around 400.

In the era of information explosion, manual processing and classification on huge amount of text data is time-consuming. It also has many operational difficulties. Besides, the accuracy of manual text classification is easily affected by human factors. A classification algorithm for news text is designed in this paper. Firstly, a large amount of news data is obtained by crawlers. They are classified into nine categories and other categories. The text features are extracted by Skip-Gram model and Word2Vec method. Then, the classification model is trained by TextCNN. The experiment shows that the algorithm in this paper can get good classification effect, the highest F1 value can reach 95.07%. For the problem of difficult classification of other category, four different classification methods are compared. The experiments show that reconstructing the data of other category can get the best classification effect.

Aiming at the problem that a large number of high-dimensional nonlinear data restrict the efficiency of combat decision making in battlefield situation, a new battlefield situation data dimensionality reduction algorithm based on local preserving discriminant projection was proposed. Based on the traditional manifold learning to ensure local information, the algorithm makes full use of the category information of battlefield situation data. In addition, intra-class divergence matrix and inter-class divergence matrix were established according to the maximum marginal criterion, so as to ensure that similar data were more clustered and dissimilar data more dispersed in the low-dimensional space after dimensionality reduction embedding, so as to improve the separability performance of situation data.

The Ripley function is a spatial point pattern analysis method, which is used to characterize point processes at different distance scales. However, to use the spatial point pattern analysis in combination with other spatial pattern analysis methods, or to analyze the spatial pattern of a population and individuals, if the spatial pattern is affected by many influencing factors, the introduction of the dimension of scale in the subsequent analysis will lead to a large number of dimensions to be examined and lack of correlation, which makes the study impossible. Therefore, in this study, the Ripley's function was improved by reducing the dimensionality of the calculated results through factor analysis, so that the results obtained from Ripley's K function only show one spatial pattern, thus eliminating the influence of scale on the spatial pattern, and expressing the spatial pattern that best reflects the current population or individual. Four standard spatial point pattern datasets were used to verify the ability of factor analysis to respond to spatial patterns after dimensionality reduction for Ripley'K, Ripley'L, and Ripley'G. The results showed that for aggregated point patterns the accuracy of the calculated results after dimensionality reduction reached 100%, where For the random point pattern, the average accuracy is also above 95%, which indicates that the applicability of factor analysis dimensionality reduction to spatial pattern models is in the random point pattern and aggregation pattern. In terms of different scale point sets, it has good adaptability to all three scale point set sizes. In this way the application scenario of Ripley function will be more extensive.

In order to solve the robustness problem of audio watermarking, this paper proposes a robust audio watermarking algorithm based on Graph Fourier transform and singular value decomposition. The Arnold transform is used to scramble and encrypt the image watermark, and then the watermark is embedded in the audio carrier by graph Fourier transform and singular value decomposition. Through experimental simulation, the audio carrier can still extract a complete watermark after various attacks such as noise, resampling, filtering, and compression, indicating that the algorithm has good robustness.

An improved ORB feature point extraction method is proposed for the disadvantages of concentrated extraction, uneven distribution of feature points, slow matching speed and high matching error rate due to the fixed threshold value set by the traditional ORB algorithm in feature point extraction. Firstly, the adaptive threshold is set by automatically calculating the greyscale value around the image block, and the feature points are judged and extracted using this threshold; the traditional ORB algorithm uses the greyscale center of mass method to determine the principal direction, but its accuracy has certain deviations, and the accuracy of calculating the principal direction is improved by Gaussian weighting the pixel greyscale values and then calculating the principal direction by the greyscale center of mass method. The experimental results show that the improved ORB algorithm has a larger number of feature points extracted, a more even distribution and a higher correct matching rate.

An efficient unsupervised deep learning image stitching algorithm using YOLOv5. The method uses a feature pyramid structure for homography and non-local blocks combined with reconstruction for image fusion. Simulation experiments show that the PSNR and SSIM of the method are significantly improved. The artifacts are significantly reduced.

In order to solve the low security and low efficiency of image encryption algorithm, this paper proposes the design and analysis of a multi-chaos fast image encryption algorithm. The algorithm combines image scrambling with diffusion, introduces Logistic and Kent to generate control parameters, and designs multiple rounds of encryption strategy related to the ciphertext. Experimental results show that the proposed algorithm has sufficient critical position and strong critical sensitivity to effectively resist attack detection. The gray scale of encrypted pixels is evenly distributed, and the correlation between neighboring pixels is weak, which can resist statistical attacks. When the number of encryption environment is roughly equal to two, the pixel change ratio and average change degree of gray value reach the best value simultaneously, and the anti-differential cryptographic analysis ability is very strong. The current image encryption size is short and real-time, which can meet the needs of real-time encryption and decryption.

For the identification of self-synchronous scrambler of error-correcting codes, a method of identification of adjacent travel ratios is proposed. Firstly, the mathematical model of the self-synchronous scrambler of error-correcting codes is given, then the biasedness of the self-synchronous scrambler of error-correcting codes is proved to be close to unbiased, secondly, the identification method of the adjacent travel ratio of error-correcting codes is given, the difference of the adjacent travel ratio of different error-correcting code unscrambler sequences is analyzed, the discriminant criterion of the maximum mean square difference fitting is established, and the identification of the generating polynomial of the self-synchronous scrambler of error-correcting codes is completed. The simulation shows that the algorithm can effectively identify the self-synchronous scrambler of linear block codes and convolutional codes and has good fault tolerance.

Blockchain technology solves the trust problem in the Internet, but there are limitations in the scalability of blockchain systems compared to traditional centralized servers, which restricts its development in practical applications. The sharding technology is a solution in the field of traditional database, which can solve the problem of limited database performance. Applying the sharding technology to blockchain system can make the blockchain system process transactions in parallel and thus improve its scalability. Therefore, studying the slicing algorithm can greatly improve the scalability of blockchain systems and promote the development and application of blockchain technology. In this paper, a random, fair, secure and stable network slicing method based on Verifiable Delay Function (VDF) and Verifiable Random Function (VRF) is proposed. Experiments show that compared with several typical blockchain system slicing algorithms, the slicing algorithm proposed in this paper has different degrees of improvement in security, throughput and storage.

The Conflict-Based Search (CBS) algorithm is one of the Multi-Agent Path Finding (MAPF) algorithms that has received much attention in recent years. It is a two-level algorithm. The high-level detects and resolves conflicts on a Constraint Tree (CT). Its low-level algorithm is to plan the optimal path for a given agent that satisfies all the constraints of the agent. In this paper, the Space-Time A* (STA*) algorithm is used as the low-level algorithm, which adds the time dimension to the A* algorithm. It means that the planning path for a single agent tends to generate and extend more and more sub-nodes. As the complexity of the problem increases, the efficiency of the low-level solution becomes an increasingly large part of the overall solution efficiency. This leads to an inefficient overall solution. To address this problem, an improved CBS algorithm is proposed. This algorithm is used when the current node in CT has conflicts. It uses current node information and Multi-value Decision Diagram (MDD) technology to pre-plan before formal re-planning. The pre-planning phase entails obtaining information pertaining to the path that satisfies the new constraint, which is referred to as pre-planning information. During the formal re-planning, the pre-planning information is used to guide the low-level algorithm to select the generated sub-nodes. The simulation experiments show that CBS with pre-planning improves the solution time by 24.4% and reduces the search space by 38.7%.

In this work, we propose a novel framework for video summarization called the Frame and Shot Network (FSNet). Unlike existing supervised video summarization methods that focus on the relationship between frames in videos, we discovered that the correlation between shots is equally significant. Our FSNet comprises of two separate modules, one for calculating frame correlations and the other for shot correlations, both of which incorporate an attention mechanism. To demonstrate the effectiveness of our method, we evaluated it on the SumMe and TVSum datasets. The results indicate that our FSNet outperforms previous state-of-the-art algorithms.

In recent years, the application of medical image semantic segmentation tasks in medical diagnosis and treatment planning has received widespread attention from the research community. The High-Resolution Network (HRNet) has good adaptability to high-resolution and high-scale medical images. In this paper, a novel high-resolution serial feature fusion encoding and decoding structure is proposed, and a CBAM attention mechanism is fused to construct a module that can jointly focus on spatial, channel, and multi-scale hierarchical information, which can improve the feature representation ability of the model and effectively reduce parameter complexity. We use the HRNet architecture to construct our model. Experimental results show that our method achieves MIoU coefficient of 98.44% on the Kvasir-SEG dataset, which is 1.43 percentage points higher than the original HR-Net model, validating the effectiveness and reliability of our method.

Our paper presents a new deep learning-based algorithm for improving the visual quality of images captured by underwater robots. The algorithm is designed to address the common issues of color distortion, low contrast, and lack of detail that often plague underwater images. By leveraging an image formation model, the proposed method is able to eliminate the effects of underwater environmental factors and enhance the color, detail, and overall visual appeal of the images. The performance of the proposed method is evaluated using objective metrics such as PSNR and SSIM, and the results demonstrate its effectiveness in improving the visual quality of underwater images. In addition, the proposed method is found to be computationally efficient, making it well-suited for use in real-time application. The proposed method has the potential to significantly improve the visual quality of underwater images and open up new opportunities for underwater exploration and conservation.

To solve the problem that traditional computer vision for elderly fall detection cannot protect the privacy of the elderly, this paper uses infrared array sensors for elderly fall detection and constructs a human infrared image fall detection system using an independently designed Raspberry Pi 4b to collect human infrared image datasets. The ECA-ResNet18 network model based on the ResNet18 network model is constructed for infrared human action recognition. Experiments were conducted on different datasets. The detection accuracy of the method reached 97.5% and 99.5% for the infrared and visible datasets, respectively, which is 5.7 and 0.8 percentage points higher than the original model; the accuracy was also improved compared with other neural network models. The results show that the ECA-ResNet18 network model has high recognition accuracy and fast detection speed in action recognition of infrared images, which has some practical application value for the promotion of intelligent pension.

Under the action of its own weight and external forces, the goaf will further induce geological disasters such as ground subsidence and subsidence, resulting in operational risks for the transmission lines attached to the surface. In order to timely grasp the deformation and settlement of the transmission lines, early warning and corresponding measures should be taken. Therefore, it is necessary to carry out real-time identification and risk early warning strategy research on the settlement risk of transmission lines. In response to this problem, this paper first uses GNSS technology and double difference model to obtain real-time settlement. Then, the cumulative trend and rate of overall and uneven settlement of different terrain towers in goaf are analyzed. Finally, the empirical estimation method is used to design a three-level risk threshold and early warning strategy to achieve rapid risk identification and early warning. The conclusions of this study can provide real-time risk identification data and targeted early warning strategies for the power industry, to assist the lean management of the power industry, to avoid and reduce the occurrence of accidents.

As a new force in the automotive industry, new energy electric vehicles are deeply loved by the broad masses of the people. As the "heart" of electric vehicles and the power core of automobile driving, the performance of the drive motor system is directly or indirectly related to travel safety and has a negative impact on individuals and society. Based on BP neural network algorithm, this paper studies the fault diagnosis of electric vehicle drive motor system and realizes the identification and judgment of fault mode of drive motor system through BP neural network algorithm, so as to improve the efficiency of fault diagnosis. And it can realize fault early warning to a certain extent, detect faults in time at the beginning of the fault, and then reduce unnecessary losses.

Considering the major problems in the earthquake risk survey, the study aims at researching, designing, and building a house drawing recognition system for the earthquake risk survey. Focusing on Optical Character Recognition technology (OCR), together with artificial intelligence algorithms such as text detection based on Differentiable Binarization (DBNet) algorithm, CRNN+CTC algorithm-based image recognition, and image segmentation with K-means algorithm, this paper elaborates on the operation process, technology selection, system design, and application effect of the system. The system shows positive effects on the collection, calibration, and output of census data, which improves the efficiency and quality of earthquake risk surveys.

With the increasing complexity of the electromagnetic environment of electronic countermeasures, the detection of radar pulse signals under low signal-to-noise ratio is becoming more and more demanding. Based on this detection performance requirement, an adaptive dynamic double-threshold signal detection algorithm is proposed based on the digital channelization and autocorrelation algorithms. The high threshold is used to detect signals with higher signal-to-noise ratios, ensuring better real-time signal detection. The low threshold is used to detect possible low signal-to-noise signals, ensuring better detection of weak signals. Simulation experiments are carried out on the algorithm, and the results show that the algorithm can effectively improve the real-time and weak signal detection ability of signal detection.

The number of enterprise application servers continues to grow rapidly, which brings trouble to the enterprise application server architecture. The complicated work makes the operation and maintenance of enterprise server inefficient. Throughout the Internet enterprises at home and abroad, the research of information automatic operation and maintenance has always been in an important position. This paper studies the fault detection of the automatic operation and maintenance platform of power company's computer room information management based on big data. Firstly, using the literature research method, this paper expounds the significance of automatic operation and maintenance of information management in the computer room of power companies and the problems existing in the platform, and introduces the related technologies of automatic operation and maintenance platform of information management. Then, by using big data mining technology, this paper studies the automatic operation and maintenance platform of power company's computer room information management, and mainly studies the accuracy of fault detection of power company's computer room information management automatic operation and maintenance platform. According to the survey results, the main types of failure problems are database failure (insufficient archiving space, cluster service exception, service exception, etc.) and virtualization platform failure (insufficient space, host performance, network exception, etc.), in which database failure accounts for about 46% of virtualization platform failure, and operating system failure (insufficient disk space, insufficient CPU memory performance, abnormal system shutdown, etc.) accounts for a low proportion of virtualization platform failure, but there are. The failure prediction of computer room information management automation operation and maintenance platform is mainly the prediction of operating system failure. Its prediction accuracy is high, but the fault prediction of database and virtualization platform is not high. It can be seen from this that the power company has a relatively large investment in power equipment, and the research on database and virtual platform technology is not mature.

With the development of science and technology and the progress of society, our living standard and quality of life have been improved. Water is the source of people's life. No matter in our daily life or in industrial production, we cannot do without water, and we also have quality requirements for water quality. The multifunctional water quality detection and purification system is mainly controlled by STM32 and integrated with TDS conductivity detection circuit, turbidity detection sensor circuit, PH value detection sensor circuit and ultrasonic sensor. The system processes the data after the analog signals of each sensor are converted into digital signals and displays each information, calculates the liquid level difference through PID, and controls and regulates the water pump valve for water purification.

Explanation is an important requirement for transparent and trustworthy recommender systems. When the recommendation model itself is not explainable, an explanation must be generated post-hoc. In contrast to traditional post-hoc explanation methods, counterfactual methods can provide scrutable and actionable explanations with high fidelity. Existing counterfactual explanation methods for recommender systems are either not generalizable or face a huge search space. In this work, we propose a reinforcement learning counterfactual explanation method MACER (Model-Agnostic Counterfactual Explanations for Recommender Systems) which generates item-based explanations for recommender systems. We embed the discrete action space into a continuous space, making it possible to use the process of finding counterfactual explanations as a task of reinforcement learning. This method treats the recommender system as a black box (model-agnostic) and has no requirement on the type of recommender system, and thus is applicable to all recommendation systems.

Most existing cross-modal hashing methods take advantage of pairwise relationships to establish correlations be-tween multimodal data. However, the majority of methods employing cross-modal hashing ignore the incompleteness and imbalance of multimodal data. To cope with this issue, we propose a Two-Step Heterogeneous Fusion hashing network for unpaired cross-modal retrieval (TSHF). In the first step, we utilize zero padding to enable image-text stitching of unpaired data and construct a triple network to learn the stitching features together. In the second step, we deploy an attention fusion network in order to explore the hidden information in the data through the inter-modal enhancement and the intra-modal complementation. In addition, manifold learning is adopted to preserve semantic similarity of original features; adversarial learning is introduced to deal imbalance in the features; label prediction ensures that discrimination is preserved. Extensive experiments on WikiData and MIRFlickr datasets indicate that TSHF has excellent performance in paired and unpaired modal retrieval.

Aiming at the problem that the power load demand and energy distribution are not balanced, leading to the continuous growth of carbon emissions and affecting the assessment effect of carbon emission reduction, a dynamic carbon emission reduction effect evaluation model for the whole life cycle of power grid is designed. The whole life cycle of power grid is analyzed and divided into fuel extraction cycle, power grid production cycle and waste recovery cycle. The wavelet denoising method is used to eliminate the noise in the carbon emission data, calculate the carbon emissions in different periods according to the carbon emission factor and the corresponding usage, determine the carbon emission reduction and carbon emission reduction rate based on the carbon emissions, and build a dynamic carbon emission reduction effect evaluation model based on this indicator, use the data collection part to collect the carbon emission data in the whole life cycle of the power grid in real time, and store it in the dynamic database, the evaluation results are displayed in the form of Web pages. The experimental results show that the carbon emission reduction and carbon emission reduction rate of the design model after carbon emission reduction are 1.1976 million tons and 16.48% respectively, and the evaluation accuracy is up to 98.19%, and the error rate under different noises is less than 0.18%, which proves that the model has good evaluation effect.

Heterogeneous Computing (HC) is a computing solution that can provide high performance and low power consumption for large-scale problems and has been successfully applied to various scenarios such as 5G communication, video processing, data mining, and edge computing. To facilitate parallel programming on multiple computing cores, the Open Computing Language (OpenCL) was created as a programming standard for heterogeneous computing systems. Unfortunately, the ReWorks operating system does not provide such a heterogeneous programming environment for developers, and thus cannot obtain the high performance offered by heterogeneous systems. This paper presents a novel OpenCL-based heterogeneous computing environment, available as a third-party library for developers using the ReWorks real-time operating system. We conducted systematic experiments to evaluate the performance of this heterogeneous computing library, and the experimental results show that programs based on the heterogeneous system exhibit superior performance compared to the original ReWorks native application.

With the development of the language service industry, computer-aided translation technology has developed rapidly. In the translation memory system, the source language text of each translation unit is stored in pairs with its corresponding target language. As the number of projects increases, there will be more and more memory files. In order to improve the operation efficiency of projects, it is necessary to make a reasonable backup and efficient management of translation memory systems. This paper takes SDL Trados Studio 2017, a computer-aided translation software, as an example to discuss the problems and solutions in the construction and application of translation memory system.

Metacognition is the critical element of personalized online autonomous learning, but it is not easy to observe or obtain. It is difficult to be monitored continuously in the practice of teaching and learning. The existing model of metacognitive ability is still in theoretical research and lacks effective model construction technology to externalize metacognition. The online learning behavior data contains rich metacognitive information. In contrast, the previous methods based on statistical analysis or traditional machine learning cannot fully extract the internal temporal and semantic features implied in the data. This study uses the self-attention mechanism and the recurrent neural network sequence model to deeply explore and analyze learners' online learning behavior and interactive text. A new evaluation model of metacognitive ability is constructed to represent learners' metacognitive ability. The research takes natural online learners' behavior data as the object to carry out experimental verification and analysis. The results show that the model's accuracy in representing metacognitive ability reaches 85.21%, which verifies the model's effectiveness.

Traditional visual Simultaneous Localization and Mapping (SLAM) is mostly based on the assumption of static environment, which is susceptible to receive dynamic targets in dynamic environment, leading to the degradation of localization accuracy. In this paper, we introduce the instance segmentation network SOLOv2, which combined with motion consistency detection can effectively eliminate the dynamic feature points in the environment and improve the visual SLAM accuracy with the depth map hole repair algorithm. Tested on the TUM dataset, the positional estimation accuracy in dynamic environments is significantly improved compared to ORB-SLAM2.

For the problem of obstacle detection of smart wheelchair on horizontal ground, an obstacle detection system was designed in this paper by using STM32 microcontroller which was based on multi-channel ultrasonic. The system mainly includes controller, ultrasonic module, LCD display module, voice alarm module and handle button module. The distance measurement accuracy of the ultrasonic module was improved with temperature compensation and digital processing of the echo signal. The RS485 bus was used to implement communication between the controller and multiple ultrasonic sensors. According to the ultrasonic information, the surrounding area of the wheelchair was divided into ten areas to distinguish the direction of obstacles. The real-time obstacle information was stored in the form of array, laying a foundation for the subsequent intelligent development. The experimental test shows that the designed system can achieve accurate distance measurement in the range of 26cm to approximately 250cm with a relative error of no more than 1%.

In recent years, due to excessive levels of indoor air pollutants in teaching, there has been an increase in the number of cases of physical illness among primary and secondary school students, thus triggering key national concerns. Traditional air purification devices mainly purify air by means of filtration, they are single-functional and work independently, which makes it difficult to manage centrally. In order to protect students' physical and mental health and improve indoor air quality, a microcontroller-based intelligent control system for space disinfection is designed. The system adopts GD32 microcontroller as the main controller to realize filtration, purification and detection functions, and the air purification equipment connects to the IoT platform through 4G CAT1 chip and MQTT protocol to realize remote detection and control. The test shows that the system is reliable and stable in operation and simple in operation, the system has multi-sensor detection function and can accurately detect ozone, ammonia, formaldehyde, PM2.5, TVOC, carbon dioxide, temperature and humidity in indoor air. Therefore, the system can effectively improve indoor air quality, it solves the problem that the current equipment on the market works alone and has a single function and has a high practical use value.

In large-scale engineering experiments, such as fluid mechanics and aerodynamics experiments, a large number of parameters and experimental data need to be recorded and managed in Computer-Aided Engineering (CAE) software by using finite element analysis to simulate physical fields. In the traditional way of dealing with tedious data, users need to repeatedly enter dozens or even hundreds of sets of data each time they conduct an experiment. In order to improve work efficiency and accuracy, recording user operations in GUI into a script file using some Domain Specific Languages (DSL) such as Python and playing that script file back to reproduce those user's operations are widely adopted by CAE software. A common script module ScriptSystem that supports recording and playback functions was introduced in this paper. It can be embedded into any CAE software, even other software that need this feature. By running scripts, those CAE software can automatically complete mesh partitioning, flow field calculation, pre-processing and post-processing. In addition, a demonstration program was developed to embed ScriptSystem into an open-source software FastCAE. The results showed that replaying scripts recorded by ScriptSystem module could obtain the same results as users do. ScriptSystem can be used as a general script module for any CAE software and other desktop software that requires recording and playback funtions. Furthermore, a message subscription module EventBus was proposed to decouple the ScriptSystem from underlying interfaces and implement the function of remote invoke.

This paper discusses the situation where the downstream retailer is the dominant player in the supply chain system. A demand function based on the price and freshness of fresh products is constructed to examine the joint fresh-keeping effort problem when the retailer and the supplier put in fresh-keeping effort separately, and to analyze the impact of the fresh-keeping effort cost sharing factor on the decisions making behavior of the retailer and the supplier. It is found that the system strategy is effective when the retailer takes up the fresh-keeping effort and when the supplier's share of the retailer's fresh-keeping effort is small; the joint fresh-keeping is not an effective system strategy when the supplier takes up the fresh-keeping effort.

With the rise of new threats represented by advanced Persistent Threat Attack (APT), traditional network security methods such as vulnerability scanning, and intrusion detection can no longer meet the actual needs. Therefore, it is increasingly necessary to make full use of more security data for Network Security Situation Prediction (NSSP) to ensure the safe operation of Internet services. In this paper, the HKIT-Trans model is effectively proposed for the NSSP task. The aim of this work is to predict the current network condition by exploiting network data characteristics. The original network data has a large number of missing null values, and it is difficult to obtain data characteristics. Therefore, we design a null filling method combining Huber regression and KNN algorithm to improve the availability of data. At the same time, we propose the improved tab-transformer algorithm to extract network data features and perform inference. The experimental results show that the proposed method has better classification performance considering the existing techniques. Compared with the same baseline model, the AUC value of our proposed null filling method is increased by 1.94%-14.91%. For the classification effect of dataset CNCERT, HKIT-Trans is 7.14%higher than Huber-KNN with Logistic Regression, and 2.26%higher than Huber-KNN with LDA.

With the rapid development of Internet information technology and wireless communication technology the demand for intelligent construction operation in production enterprises is increasing, and the requirements of wireless positioning technology are becoming more and more stringent, but due to the influence of the complex indoor environment, the outdoor positioning system has been unable to meet the needs of indoor positioning. At present, the relatively mature indoor positioning technologies include WI-FI, Bluetooth, infrared, RFID and so on, which have a common disadvantage: large positioning error. Therefore, in this paper, combined with the problems existing in the on-site construction of production enterprises, taking stm32 as the main control chip, using UWB communication and positioning technology, independently designed PCB, developed an intelligent monitoring system based on UWB positioning technology, which can not only meet the outdoor high-precision positioning, but also achieve indoor centimeter positioning. At the same time, the system is equipped with voice, video and other modules, combined with the terminal display platform to realize the functions of mobile track display, voice interaction, image transmission and so on, which is convenient for managers and wearers to use together. To ensure real-time monitoring of the construction process and timely handling of emergencies.

In this paper, a decentralized stabilization scheme of linear time-varying large-scale interconnected systems with unmeasurable states is proposed. These interactions among subsystems are also time-varying and affect each subsystem through its input. First a decentralized observer scheme for time-varying interconnected system is constructed by the dual optimal control solution to obtain good estimations for the unmeasurable states. Based on these observer states, a time-varying decentralized feedback law is introduced to achieve the global system exponential stability. The solutions of time varying Riccati equations are obtained by backward Euler’s method and implemented with the original system dynamics. Computer simulations of the responses of an example are also conducted to show the effectiveness of the proposed approach.

The evaluation of aircraft penetration effect has always been an important aspect in the field of military confrontation. This paper comprehensively analyses the force situation of the aircraft, calculates the flight trajectory of the aircraft and the jammer accurately through simulation, and analyses the influence of the release time, release speed and release angle of the jammer on the shielding effect in combination with the radar jamming mechanism. The neural network is trained with a large amount of simulated data, and the optimal release decision under certain scenario is calculated offline with particle swarm optimization algorithm.

As the terminal network of the power system, the distribution network has complex power supply structure, and the traditional power grid error prevention strategies are not so suitable for direct use in the distribution network. Based on network topology analysis, intelligent control, intelligent lock control and other technologies, this paper proposes a set of intelligent error prevention system for distribution network, which can effectively guarantee the safety of distribution network operators and improve the safe and stable operation level of distribution network.

Motion target detection is a prerequisite for road monitoring, motion target tracking, instance segmentation and other tasks. UAV video images are easily affected by some unavoidable factors in the acquisition process, such as wind interference and own motion during the shooting process can lead to image background changes, target scale changes and intermittent motion, making the motion target detection task more challenging. To address the problems of poor accuracy of existing UAV video motion target detection methods based on deep optical flow networks and the limitation of target detection performance in complex scenes due to the complex and diverse features of UAV video data, this paper proposes a new UAV video motion target detection method based on optical flow networks. Firstly, a convolutional structure reparameterization method is used in the coding part to further fuse detailed and semantic information to improve the feature expression capability of video images; secondly, the self-attentive global motion feature enhancement module proposed in this paper is introduced to improve the network's ability to extract global information and better combine contextual information to achieve more accurate optical flow estimation; finally, the optical flow threshold segmentation is used to obtain different motion target detection results for different scenes by optical flow threshold segmentation. In this paper, three sets of low-altitude UAV video data from different scenes are selected for experiments on the public dataset AU-AIR2019, and the experimental results prove that the proposed method can achieve better motion target detection results in single-target, multi-target and occluded target scenes, and it is better than the current mainstream optical flow networks: FlowNet1, PWC-Net, HD3, PWC-Net and HD3 on the public dataset FlyingChairs. PWC-Net, HD3, GMA metrics EPE (end point error, EPE) on the public dataset FlyingChairs, and improves the RAFT by 0.10 over the benchmark network in this paper, which effectively improves the accuracy of UAV video motion target detection by deep optical flow networks.

Estrogen Receptor is closely linked to the growth of breast cancer, and breast tumor cells significantly depend on Estrogen Receptor. Therefore, it is of great practical significance to find compounds that can antagonize ERα activity as candidate drugs for the treatment of breast cancer. In drug research, the method of establishing compound activity prediction model is usually used to screen potential active compounds. In this paper, according to the provided ERα antagonist information, based on the screened biologically active compounds, the ADMET properties of the compounds were predicted by three different classification models. Predictive modeling studies were carried out; finally, the value range of each element in ADMET properties was imprecisely optimized to better guide the selection of anti-breast cancer drug candidates.

For areas with limited water resources, seawater desalination is widely used as a solution for water supply, especially in coastal cities. In order to predict the water quality after desalination, this paper proposes a water quality prediction model based on the classification algorithm of random forest. Factors such as pH, hardness, solids and so on are selected and used to build this water quality prediction model. Compared with the prediction accuracy of logistic regression model, the result shows that the water quality prediction model established by random forest classification algorithm has a slightly higher accuracy. Random forest algorithm has the characteristics of high classification accuracy, good practicability, and can effectively monitor water quality.

In recent years, the concept of edge computing has risen. The network computing structure has gradually changed from a centralized cloud service cluster to a cloud-edge collaborative service system. But at the same time, the limited resources of edge nodes and the complex network environment make it easier to become the target of network intrusion, resulting in the collapse of nodes. Aiming at the security threats faced by edge nodes, we propose an edge network intrusion detection method based on feature selection and TextCNN. Enhance the correlation between features and eigenvalues through feature selection, and reduce the intrusion detection cost of nodes. The feature information of the sample is extracted through TextCNN, and the feature information is combined and correlated through the proposed correlation pooling layer, so as to improve the accuracy of model detection. Finally, the method is tested on the kddcup99 dataset. Compared with traditional methods, the accuracy, recall and f1 score of the model are improved, which proves the feasibility and effectiveness of the method.

Classifier based on deep learning has greatly improved the ability of identifying malware. However, these detectors may be affected by adversarial input disturbances. Any vulnerability in the malware detector may pose a significant threat to the platform they protected. Therefore, in order to improve the defense against malware and adversarial examples, we conducted the study on adversarial deep learning for Android malware. On this basis, this paper proposes a new method to improve the robustness of the malware detector which is based on visualizing and deep learning. Different from the traditional method with feature extraction, this paper first converts the malware binaries of the Android application package into grayscale images and use the Generative Adversarial Network (GAN) to optimize the boundary samples. Then conduct adversarial training to generate a robust malware detector based on deep learning. A large number of experiments have proved that this method can resist different attacks without occupying too many resources. The results show that it has significant advantages compared with other current classifiers.

As the number of users using electric vehicles continues to increase, the amount of user charging behavior data in the electric vehicle smart charging platform has also shown an explosive growth trend. By profiling users' daily charging behaviors, it is used to detect whether users' charging behaviors are abnormal and prevent the generation of behaviors such as electricity theft. In this paper, we propose a method to construct a user abnormal behavior detection model based on a smart charging platform. The aggregation validity index is constructed and used to determine the optimal classification number K value of the K-means clustering algorithm, the optimal set of user features is extracted by the redundant dynamic weight feature selection algorithm, the abnormality threshold is set, and finally the user abnormal behavior model is constructed based on softmax regression. Finally, the effectiveness of the method is demonstrated by comparison analysis.

In complex industrial scenarios, multiple networks and different protocols exist, and communication between devices is difficult. To solve this problem, we looked at Modbus, OPC, MQTT protocols, and API interfaces. Based on software definition technology, the digital device model of controller, data pump and connection pool is constructed. The controller collects the original data. The data pump processes the collected data according to the rules and can trigger notification or alarm. Connector connects to a gateway or database. Developed the industrial Internet heterogeneous network integration management software, with driver management, connection management, event management, user management, configuration management and other functions. The system realizes the integration of heterogeneous network and edge calculation of industrial Internet and has been applied in engineering and obtained good results.

The recommendation system aims to obtain valuable information for users and alleviate information overload. Graph Neural Network (GNN) is one the mainstream method of recommendation systems for its powerful capabilities of graph data representation and deep feature extraction. But there are still problems with the efficiency and accuracy of GNNs. Therefore, a High-Efficiency Graph Neural Network (HEGNN) is proposed in this paper to build a lightweight graph recommendation system. HEGNN strengthens the local and global preferences of users with attention blocks. It abandons the feature transformation and nonlinear activation layer of vanilla GNNs. Only the basic components are reserved to improve efficiency. Comprehensive comparative experiments with nine baseline algorithms are carried out on three benchmark datasets which include Amazon-Book Dataset, Yelp2018 Dataset, and Gowalla Dataset. Compared with existing recommendation methods such as NGCF and LightGCN, HEGCN not only achieves the highest score on two evaluation metrics of Normalized Discounted Cumulative Gain and Recall but also requires the least training time.

As a strip building laid on the ground, highways are prone to road hazards such as pavement cracks, potholes, subsidence, tension cracks, extrusion bumps, and distortion deformation with the increase of service life. If these hazards are not detected and handled in a timely manner, they will cause significant traffic safety hazards. Therefore, regular road hazard detection is particularly important. Traditional road hazard detection methods are manual detection or detection with road detection vehicles, but due to the limitations of detection methods and processing technologies, they cannot meet the real-time requirements for road surface detection. Automatic road surface defect detection methods have proved their effectiveness in traditional road surface detection, and their real-time, fast, and accurate detection capabilities have established their potential advantages in automatic detection methods for road surface defects. This paper comprehensively analyzes the automatic detection and processing technology of road surface defects, explores different dimensions of automatic detection and processing technology for road surface defects, conducts detailed comparative analysis of road surface defect detection technology based on deep learning, Overall, the above work can promote the development of road surface defect technology and provide reference and basis for future road hazard detection methods.

With increasing number of software vulnerabilities, the quantity of attacks utilizing malicious samples is also on rise, leading to intensified adversarial competition. In particular, the application of automatic vulnerability mining techniques has resulted in a significant increase in the number of vulnerabilities, but security researchers often do not have enough time to deal with them. This paper proposes a symbol-execution-based automated vulnerability exploitation method, which can achieve automation of vulnerability detection, classification and exploitation. Finally, this paper designs and implements a prototype system for symbol-execution-based automated vulnerability exploitation and verifies its effectiveness through experiments. This research provides security analysts with an in-depth understanding of the types of vulnerabilities and determines methods for vulnerability exploitability, further improving the efficiency of analyzing and fixing vulnerabilities.

Unmanned Aerial Vehicle Ad-hoc Network (UANET) are gaining extensive attention in fire monitoring, communication relay and other fields. Because of mobility of UAVs, network topology changes frequently. In OLSR routing protocol, each node broadcasts HELLO packets at regular intervals for link sensing and neighborhood detection. However, if the HELLO interval is too small when the node speed is slow, unnecessary traffic will appear in the network. If the HELLO interval is too large when the node speed is fast, the performance will be degraded too. This paper proposes a routing protocol that adaptively adjusts the HELLO interval. Large amount of simulation results of NS-3 is used as samples, the neural network is trained by GA-BP (Genetic Algorithm, Back Propagation) algorithm, and the chosen network performance metrics under different HELLO intervals are predicted according to the speed of nodes. The MADM (Multiple Attribute Decision Making) method is used to comprehensively evaluate these metrics and select the optimal HELLO interval. Our simulation results show that compared with the original OLSR and other schemes, the proposed scheme can achieve a large performance improvement at a very small cost.

In statistics, the evaluation of measurement uncertainty is used as the technical carrier for the quality evaluation of measurement results of automatic testing equipment. In this paper, the uncertainty of the measurement results of the automatic testing system of Twin-wheel side force coefficient, which is an important testing equipment for pavement lateral force coefficient, is evaluated according to the testing methods specified in the Chinese transportation industry standards. The sources of uncertainty are analyzed, and the calculation steps of uncertainty combination are simplified. The purpose of this paper is to solve the scientific problems such as insufficient basic data and heavy calculation encountered by engineering inspectors in the evaluation of uncertainty, and to provide support for the evaluation of road traffic engineering inspection quality.

This paper proposes to analyze the research and application status of LoRa through bibliometry and the knowledge map generated by CiteSpace, and to clarify its research hotspot and development trend. For methods, the 2000-2022 in the National Knowledge Infrastructure database (CNKI) is used. The Excel was used to generate annual distribution, CiteSpace was used to generate knowledge maps of authors, institutions, keywords, etc., and analyze the research application status and frontier of LoRa. A list of results includes a total of 2209 effective documents were obtained, and the number of articles is the author and universities, close cooperation and small cooperation; the nature of LoRa, low power applications include environmental monitoring, intelligent water meter, etc.; long distance applications include data acquisition, data transmission, wireless communication, wireless transmission, etc.; LoRa is more and more widely used, but not very deep, and does not last long. Based on the analysis results, it is found that the research application of LoRa is becoming more and more extensive. From the transmission of a small amount of data to the monitoring and transmission of a large amount of multidimensional data, it can be reasonably inferred that the application of intelligence is still the next research hotspot, and LoRa will still be the main tool in intelligence.

With the continuous development of information technology, artificial intelligence fire automatic alarm system is gradually developed. However, the current system is based on a single sensor for identification, when a fire still uses manual active fire extinguishing, low efficiency in handling fire, high failure rate, poor system scalability, poor anti-interference ability, high false alarm rate. At the same time, the recognition accuracy is low, the processing speed and intelligence are also insufficient. Therefore. To address the above problems, this study designs a multi-sensor artificial intelligence fire prevention system using smoke and temperature sensors. The fire prevention system can send alarm signals at the earliest stage of fire and can intelligently extinguish the fire, so that the disaster can be controlled and extinguished by taking measures in advance. Meanwhile, the system can continue to work even if one of the sensors is not working properly.

Recently, the development of medical imaging technology has made computer image analysis methods an indispensable tool for clinical diagnosis in the medical field. Among these methods, Magnetic Resonance Imaging (MRI) technology provides doctors with a range of anatomical images, helping them locate lesions quickly and accurately. In this article, we propose a nested network named U2-SegNet that could be trained from scratch without relying on pre-trained networks. This architecture can effectively address segmentation challenges by capturing multi-scale information. Our approach leverages the following features: (1) a well-designed residual U-block (RSU) that captures contextual information of varying scales by mixing receptive fields of different sizes, (2) full utilization of multimodal MRI data to address data scarcity, (3) incorporation of Salient Object Detection (SOD) to enhance global and local contrast information, leading to smoother segmentation edges. We achieved impressive results on the BraTS18 dataset with our proposed model.

In order to meet the requirements for 4K ultra high-definition video recording, storage and playback, as well as the need for long-term capture and recording of screen information of the operating terminal in important control scenarios such as shipborne, on-board and airborne, it is convenient for post event technical analysis, service quality assessment, drill rehearsal and operation accident liability determination. The domestic video codec processor Hisilicon HI3531D is studied. A video recording and playback system supporting 4K video capture and H.264/H.265 codec is designed and implemented. The compression algorithm of H.264/H.265 coding standard is used to realize the compression, storage and functions of video data. Single channel codec 3840x2160@30fps resolution can reach 17.6Mbps, the codec system delay can reach 110ms in JARI Works OS, and the dual redundant Ethernet switching time can reach 23.5ms.

Today, there has been a great shift in the way data is utilized, and data has risen to become a new factor of production and a new driving force for business development in the digital era. The commercial application of new-generation information technology has reduced the cost of data processing, accelerated the accumulation of data resources, and given rise to new business development models. These enterprises use data as "raw materials" and digital technology as means to provide big data products and services to the outside world, which are called big data service-oriented enterprises. With the recognition of the value of data by the whole society and the emphasis on digital transformation, the big data service industry is in an accelerated expansion stage. On the contrary, in academia, there is still a gap in the research on the value of big data service-based enterprises. The special nature of the business model of big data service enterprises and the diversity of value influencing factors cause uncertainty in the future earnings of the enterprises, comparable companies are difficult to find, the reports cannot reflect the overall value of the enterprises, and the application of traditional valuation methods is limited. In this context, this paper examines the issue of valuing big data service companies.