This PDF file contains the front matter associated with SPIE Proceedings Volume 12941, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

For the determination of whether a ship's flight segment was threatened in the course of ship-to-air defensive operations, the military requirements were analyzed, and the establishment of an algorithm coordinate system and the classification of the relationship between the segment and the enemy's threat range were carried out. On this basis, the determination rules and algorithm models in each case were studied and established, then an example was given for calculation, analysis, and verification. The verification results showed that the judgment method proposed in this paper can quickly and accurately determine whether the segment on the shipping route is under air threat, which would provide important and practical decision support for the research and formation of scientific and reasonable navigation plans.

For the problem is existing that a single model could not get high precision tracking of maneuvering targets in complex air routes, a ballistic trajectory matching method using combined models based on current statistical models and S maneuver models is put forward. Moreover, the idea of using α−β−γ variable gain filter waves to reduce algorithm complexity is given to avoid the deficiencies of conventional Kalman filtering algorithm and particle filtering algorithm and solve the problem of lack of engineering. Finally, this algorithm is verified by simulation tests for its effectiveness.

As an essential support for technology applications such as cloud computing and artificial intelligence, data centers face a pressing need to balance their demand with their own high energy consumption. Since the energy consumption of the Information Technology (IT) systems and cooling systems accounts for the majority of total energy consumption in data centers, a collaborative optimization framework called COHPPO based on a Hybrid Proximal Policy Optimization algorithm (HPPO) is proposed, which controls IT system and cooling system in a coordinated manner, fully compressing the resource redundancy of the IT system and cooling redundancy of the cooling system. The HPPO algorithm is trained and verified using a task dataset and a simulation platform that simulates the environmental state inside the data center. The experimental results demonstrate that the COHPPO collaborative optimization framework proposed in this paper can effectively improve the energy efficiency of the data center and the stability of its internal environment.

In order to adapt to marine operations and improve the positioning accuracy of ships, this paper proposes an improved marine predator's algorithm for ship power system thrust distribution. An objective function for minimizing the power loss of the propulsion system is established. The function is constrained by the angle, magnitude and rate of change of the thrust direction. To solve the problem that the conventional Marine Predator's Algorithm (MPA) easily falls into local optimum, an adaptive t-distribution change operator is introduced to update the population to avoid falling into local optimum. The updated populations are grouped and learned according to the fitness to further improve the population quality and search accuracy. The traditional marine predator's algorithm is also compared with the improved simulation results, and it is concluded that the improved algorithm can deliver the commands of the power positioning system controller more accurately under the environmental disturbance and effectively reduce the power loss when the ship is sailing.

The flight management system is the core module of the avionics system of large civil aircraft, and lateral navigation is one of its important functions, which can assist the pilot in realizing the fully automatic flight of the aircraft and improving the economy and safety of the aircraft flight. In this paper, a model of large civil aircraft dynamics is studied, and the model of lateral navigation guidance algorithm is built in MATLAB/Simulink. The model use PID, cascade PID, and cascade fuzzy PID controllers to steer the aircraft towards the direction of reducing cross track distance respectively, and finally it is verified in the flight management semi-physical simulation system. The result demonstrates that the lateral navigation function algorithm can complete the navigation task and meet the flight requirements of RNP0.1, which has practical engineering applications.

Urban rail transit passengers have significant heterogeneity, and their travel behaviors are complex and diverse. In the study, passengers cannot be simply analyzed as a group. Passenger classification is an important method to study the similar behavior and rules of passenger groups. At present, many scholars at home and abroad focus on the analysis and classification of passenger travel characteristics based on urban traffic big data. In this context, it is necessary to comprehensively analyze the travel behavior mechanism of individual passengers, and gradually complete the fine classification of passengers, so as to solve the problem that the existing models and algorithms are not sensitive to the heterogeneity of individual passengers. Therefore, based on the multi-dimensional travel characteristics of individual passengers, this paper comprehensively applies the two-step clustering and fuzzy C-means fusion algorithm to build a hierarchical passenger clustering model, so as to realize the fine classification of individual passengers. Finally, taking the multi-dimensional travel characteristic index set of Suzhou rail transit passengers as an example, the fine classification process of individual passengers is successfully realized, and the classification effect is good.

At present, China is in the stage of aging over age, as the aging of more serious, Alzheimer's disease (AD) prevalence increased year by year. Alzheimer's disease trend prediction algorithm prediction accuracy is not high, prediction time is short, combining with regression algorithm model for long-term prediction method. The regression algorithm can predict continuous variables, which is conducive to reveal the pathological mechanism of Alzheimer's disease and can more accurately realize the auxiliary diagnosis and prognosis prediction of Alzheimer's disease. With the development of medical imaging technology, the application of regression algorithm to the clinical score of Alzheimer's disease is increasing. In this paper, deep learning method is used to predict Alzheimer disease onset trend based on regression algorithm model.

This paper proposes an improved Astar algorithm based on domain expansion strategy to address the problem of traditional Astar algorithms that plan paths that are not globally optimal and have large path lengths. Specifically, expanding the traditional Astar algorithm's eight search domains to 24 domains, increasing the search scope; optimizing the heuristic function by using a dynamic weight heuristic function instead of the traditional one. In order to verify the effectiveness of the improved Astar algorithm, simulation experiments were conducted on both the traditional Astar algorithm and the improved Astar algorithms in the same environment. The simulation results show that compared to the traditional Astar algorithm, the improved Astar algorithm based on domain expansion strategy planned has a shorter planned path, lower total path cost, and better overall search performance.

The flexible job-shop scheduling problem surpasses the limitations of conventional workshop scheduling problems that reduce machine constraints, increase uncertainty, and belong to the NP-hard problem. An adaptive simulated annealing genetic algorithm based on reinforcement learning is put forward to overcome the constraints of complex parameter determination and poor local search capabilities that standard genetic algorithms face when dealing with flexible job shop scheduling. The approach uses a multi-parent POX crossover operation and introduces a simulated annealing algorithm into the mutation operation to enhance the method's capability for both global and local optimization in the evolution process. The tournament method is combined with the optimal strategy to ensure the algorithm's convergence. The crossover and mutation parameters are dynamically adjusted and optimized using the reinforcement learning algorithm in conjunction with the improved simulated annealing genetic algorithm so that the parameters of the algorithm can adapt to the evolution process according to experience, and the searchability and computational efficiency of the algorithm are improved. By testing and examining the common examples, the proposed algorithm's rationality and superiority are ultimately demonstrated.

In view of the problems of fresh food distribution, this paper analyzes the actual situation of fresh food distribution, selects total cost minimization and customer satisfaction maximization as optimization objectives, then adds time windows and other constraints to construct a dual-objective reefer truck path planning model, and uses genetic algorithm combined with greedy algorithm to solve. The experimental simulation results show that the algorithm of the article has good planning effect in path planning.

The current analysis model of the causes of cases is generally single objective, and the analysis scope of the causes is small, leading to the increase of analysis deviation. For this reason, a method for constructing a case cause analysis model based on the decision tree algorithm is proposed. According to the current measurement requirements and standards, first determine the source of case cause data, and then build the project of case cause characteristics through pre-processing of case cause data to achieve basic data preparation. Next, design the case cause analysis model, build a multi-level model analysis structure by means of training and evaluation, expand the actual analysis scope, and complete the model construction and ability enhancement. The final test results show that: after three stages of testing and research, the causes of four randomly selected cases were analyzed, and the final real-time analysis deviation value of the model was better controlled below 0.25, indicating that with the aid and support of the decision tree algorithm, the analysis accuracy of the causes of cases has increased, the application ability of the model is better, and has practical application value.

The Dynamic Facility Layout Problem (DFLP) is the problem of arranging facilities of different sizes in a workshop over multiple planning periods in order to meet some objectives and constraints. In this article, we consider the two objectives of minimizing both the material handling fees and facility rearrangement costs and propose a Multi-Objective Ant Colony Optimization Algorithm (MOACO) based on Flexible Bay Structure (FBS). The algorithm uses the FBS for coding and combines global search based on niche technology with Pareto optimization based on local pheromones to search for Pareto-optimal solutions. To improve the quality and variety of solutions, a paired exchange method and five layout update techniques are put forward. The MOACO algorithm is tested on three typical dynamic instances, and the results of the experiment demonstrate that the proposed algorithm is an efficient method for the DFLP.

Aiming at the problem of difficult accurate differentiation between public transportation and private cars in current research, a station matching algorithm based on mobile phone signaling data is proposed to address issues such as low frequency and unstable collection frequency of mobile phone signaling data, and low positioning accuracy. The algorithm utilizes the average speed between neighboring base stations that the user's phone is connected to during travel to infer the traffic status of a road segment. By comparing the average speed of the road segment with the median average speed of the entire trajectory, the algorithm determines whether the road segment contains important traffic nodes, such as bus stops or intersections, and calculates the station density of the entire trajectory to further verify the results predicted by the existing classification model. Experimental results show that the algorithm can significantly improve the identification accuracy of buses and private cars. Compared with using machine classification methods alone, the accuracy has increased by 11%, reaching 94%, which can provide strong support for the collection and analysis of traffic information.

In the process of outdoor positioning and navigation of the patrol robot dog, the collected information was not fused, which led to the failure of accurate identification of obstacles. Therefore, an outdoor positioning and navigation algorithm based on multi-sensor fusion was proposed for the patrol robot dog. The information was collected by multi-sensor equipment such as 3D laser radar erected on the section plane of the patrol robot dog. The information was decomposed based on the wavelet packet energy spectrum strategy. After the robot dog patrol target was detected, the fusion results are input into the extended Kalman filter model through the Bayesian initial fusion data, and the robot dog positioning is realized according to the optimal variance estimation of the output parameters. Finally, the ant colony algorithm is improved by the clustering search strategy to search the optimal path to achieve navigation. The experimental results show that the method can effectively identify obstacles in complex environments and different scene execution speeds, and the recognition accuracy is high; It can accurately and quickly plan the best path of the fault area to ensure the stability of power system operation.

Based on the adaptive finite element algorithm of the numerical model, the mesh is automatically coarsened by element merging for the grid structure of the quadrilateral, which aims to simplify the simulation model and improve the simulation calculation speed. First, the fine mesh is directly converted into pure geometric elements according to the element nodes, and then the mesh elements are redivided according to the boundary hard points and internal geometric hard lines and mesh scale. Extract the geometric boundaries of each different attribute subregion that needs to be coarsened, and then smooth the geometric boundaries and select the key boundary nodes to form the shape boundaries of the coarse mesh. The polygon area formed by these boundary nodes is remeshed to make the mesh size meet the coarsening requirements. According to the above ideas, this paper finally gives the specific program flow of coarsening on the quadrilateral grid.

The continuous development of network accounting has promoted the continuous improvement of accounting operation means, objectives, functions, and supervision and management of enterprises. However, it cannot be ignored that network technology has also brought serious security threats to accounting information system. This article mainly conducts research on network accounting information security based on classification and regression tree algorithm. This article first introduces the security risks of network accounting information, then analyzes the principle of decision trees and the selection of classifiers. Finally, this article uses classification and regression trees to construct a network accounting information security model. Through the experimental results, it can be concluded that the network accounting information security model has a recognition rate of 90.33% for accounting information, which is superior to most information security models.

The optimal MPR (Multi Point Relay) algorithm is a means of addressing the drawbacks, such as high overhead, faced by the OLSR (Optimized Link State Routing) protocol in link state routing protocols. However, traditional MPR algorithms suffer from redundancy issues, leading to suboptimal solutions in certain scenarios. To tackle this problem, this paper considers solving the optimal MPR set using a genetic algorithm approach. Firstly, the traditional ant colony algorithm is improved, and the maximum-minimum ant colony algorithm is proposed for MPR. To reduce algorithm complexity and lower time costs, a fruit fly inspired optimal MPR set selection algorithm is introduced. Simulation results demonstrate that this algorithm can avoid redundancy issues found in traditional algorithms and has a shorter running time compared to the ant colony algorithm-based approach.

With the rapid development of the manufacturing industry, the design of manufacturing processes has become increasingly complex, and there is a growing demand from companies and users for efficient process recommendation systems. However, existing research on process recommendation often overlooks the importance of user intent. In reality, user intent often consists of multiple facets, each with different emphases, collectively driving users to choose different processes. For example, an intent to meet tight deadlines may favor less time-consuming processes, while an intent for safety requirements may prefer more mature processes. Disregarding the notion of multiple intents from users would confine the modeling of interactions between users and processes. Therefore, to enhance the efficiency and quality of process design, this paper proposes an intent-aware manufacturing process recommendation algorithm. This algorithm combines process knowledge activity analysis with intent awareness and establishes a mapping between process knowledge and process knowledge graph through graph representation learning and attention mechanisms, enabling perception of user intent and corresponding process knowledge recommendation. Through empirical validation and application, this paper demonstrates the effectiveness and practicality of the algorithm. Experimental results show that the intent-aware process knowledge recommendation algorithm significantly improves the efficiency and accuracy of process design, providing strong support for the development of the manufacturing industry.

Considering that the traditional bonded fuel supply chain has problems such as weak trust relationship and low cooperation efficiency, which affect the profits of supply chain node enterprises. To this end, a two-level supply chain game model based on suppliers and distributors is built. First, the particle swarm optimization algorithm is used to estimate the supplier's profit function, and the initial game strategy of the supplier is obtained accordingly. Add the neural network to the ant swarm intelligence algorithm to make it optimize the operation, and finally realize the optimal matching of supplier and distributor business. Finally, the effectiveness of the proposed method for improving the collaborative optimization of the bonded fuel supply chain is verified by experiments.

Target detection is a key computer vision technology that is utilized extensively in crucial sectors including aerospace and aviation as well as fine-grained security detection. One of the most frequent problems in computer vision is the implementation of adversarial training in vision algorithms. The target detection technique we present in this research is an adversarial machine learning-based approach that uses GANs. The images are inputted and sent to the GAN network and the detection network, respectively, and then the images produced by the GAN learning are blended with the real images and sent to the target detection network. This algorithm uses GAN as an adversarial trainer for the baseline target detection algorithm. Finally, YOLOv5s is selected for experimental validation, and the data of VOC hybrid 2007 and 2012 are selected for experiment, and the data in the Based on YOLOv5s, its map0.5 and map0.95 have a good improvement in detection effect compared to the original YOLOv5s, and finally based on the comparison experiments and ablation experiments, it shows that the GAN-based target detection algorithm out of this paper is effective.

Based on the intelligent optimization algorithm, this paper attempts to solve the complex contact stator guide curves of curve hydraulic motor. Use the Gold Algorithm (GJO) to improve the particle group algorithm (APSO) that is easy to fall into the optimal problem, and a more advantageous Golden Jackal Particle Swarm Algorithm (APSO-GJO) is obtained. By comparing the global and local optimal solution accuracy levels and convergence speeds of the three algorithms, it is proved that the improved particle swarm algorithm has excellent fast convergence and the ability to find the global optimal solution, which has a significant effect on the optimization of the contour guide curve of the equal contact stator.

SQL injection attacks are pernicious forms of cyber assaults, and the integration of the TF-IDF algorithm into the domain of SQL injection detection has emerged as a prevailing trend. To address the shortcomings of traditional TF-IDF algorithms, which neglect feature distribution and insufficiently extract features, this paper proposes a detection method for SQL injection attacks based on the TF-IDF-CHI algorithm. This algorithm not only remedies the inadequacies of the TF-IDF algorithm in terms of feature distribution but also enhances feature extraction by incorporating category factors and an improved CHI statistical approach. Experimental findings substantiate an approximate 5% increase in precision compared to the traditional TF-IDF algorithm, thus underscoring the superior performance and efficacy of the proposed algorithm in detecting SQL injection attacks.

The 0-1knapsack problem is well known to be NP-hard in optimization algorithms. There are many commonly used methods to optimize such problems, such as dynamic programming, enumeration, pruning and bounding, greedy algorithms, backtracking, etc. With the advent of some modern heuristics, such as swarm intelligent optimization algorithms, more options have been provided for solving such problems. The Gene-tic algorithm is one of the most commonly used algorithms. The idea of this paper is as follows: firstly, the mathematical model of 0-1 knapsack problem is given, combining the characteristics of the model, the idea of double structure is introduced, combined with the greedy algorithm, the two-step method of value factor is proposed, and the secondary selection coding is carried out. In the selection operation, recombination is used to improve the convergence of the following algorithm. Compared with the traditional binary genetic algorithm, the genetic algorithm with penalty terms and the Standard Genetic Algorithm (SGA), the convergence of the improved binary genetic algorithm is significantly improved.

This paper summarizes the development status, prediction methods and classification of wind power prediction at home and abroad and verifies three typical power prediction models through example comparison- -ARMA power prediction model, Kalman filter model and a distributed solar power generation area. Through the comprehensive analysis of the evaluation indicators of the prediction results, it is found that the wavelet neural network prediction model has the highest accuracy and realizes the short term and super short term accurate prediction for the small range of distributed photovoltaic and wind power. To meet the actual operation needs of the power grid, it has a certain promotion and application value in the distributed power generation projects.

This paper describes a tree-based decomposition and subexpression exchange method for generating new strategically valid handwritten mathematical expressions (HMEs). This approach uses the Edmonds algorithm to extract a Minimum Spanning Tree (MST) from a directed view and groups input strokes to segment symbols and label spatial relationships between symbols and symbol pairs. By training the HME recognition model and the mathematical Language Model (LM), we aim to improve the recognition performance and generalization ability of existing state-of-the-art methods from both data enhancement and model design perspectives. By adding HMEs of different styles, complexity and difficulty levels in the training dataset in order to increase the adaptability of the system to various forms of expressions.

This paper aims to research the feature extraction method for bearing faults based on the wavelet packet algorithm. Rolling bearings, as crucial components of rotating machinery, play a vital role in diagnosing faults to enhance the reliability of such machinery. With the development of artificial intelligence technology, the wavelet packet algorithm has gradually demonstrated its significance in industrial production. In this study, the wavelet packet algorithm is applied to extract features from bearing fault signals, and different types of fault signals (such as inner race faults, rolling element faults, and outer race faults) are analyzed in detail. By extracting energy values, significant changes in energy spectra between normal and abnormal signals can be effectively identified. This wavelet packet energy spectrum-based bearing fault detection method overcomes the limitations of single singular point for bearing fault detection and validates the effectiveness of wavelet packet energy spectrum as a feature to distinguish normal signals from fault signals. The obtained feature values also provide beneficial data analysis basis for further fault detection, offering practical significance and application value in bearing fault prediction and maintenance.

The application of deep learning techniques significantly improves the performance of target detection algorithms. The article provides a comprehensive discussion of target detection principles, methods, algorithms, and future trends. First, the principles of target detection algorithms are summarized, including tasks, evaluation indexes, and public datasets. The image-based target detection algorithms are divided into two levels: traditional target detection and deep learning-based target detection. A brief overview of the types and shortcomings of traditional target detection algorithms is provided. Deep learning-based target detection is divided into single-stage and dual-stage for specific description. It summarizes the mechanisms, advantages, and limitations of each type of algorithm, conducts a comparative analysis, and suggests the future research direction of deep learning-based target detection.

In order to realize intelligent recognition and statistics of hydrogenation catalyst electron microscope image information and speed up electron microscope experiments. This paper proposes an improved CenterNet hydrogenation catalyst target detection algorithm, which uses Resnet50 backbone network to extract features, and adopts two sets of spatially separable convolution optimization residual modules of different scales to improve feature extraction capability. Moreover, FPN (Feature Pyramid Networks) is added to integrate shallow and deep features. In order to further solve the problem of background interference, the coordinate attention module is introduced to adjust the weight from adaptation, which can highlight the target characteristics of the catalyst more effectively. At the same time, the backbone network is optimized by referring to DenseNet in the downsampling stage, and on this basis, an improved ASPP module is added to adjust the sampling rate and increase the receptive field, so as to further improve the network's ability to identify and locate the target. Experimental results show that the AP value of the improved CenterNet model improved by 19.37, 20.16, 24.80, and 12.92 percentage points over the original CenterNet, YOLOv4, SSD, and Faster R-CNN detection algorithms in the same experimental setting. After the improvement, the detection accuracy of the model is obviously improved, and the model has a high capability of hydrogenation catalyst target detection.

Human key point detection has essential application prospects in human and computer interaction and disease prediction. However, in the current scenario, there is still some room for improvement in the accuracy of key point detection when facing complex scenes and some key points are missing. To address the above situation this paper proposes an improved YOLOv8-SW model based on the YOLOV8 model, adding the SA attention mechanism to improve the model's ability to extract channel and spatial features, replacing IOU with WIOU, and increasing the proportion of the model's weight on the general quality box. The validation results on the COCO2017 dataset show that the improved model can achieve better detection performance for complex scenes. Compared with the baseline method, the accuracy is improved by 1.0%, which is significant in promoting the practical application of human key point detection.

ATC system is a key functional component of machining center, and its reliability level is very important to the machining center. ATC system has multiple fault modes. In order to construct its reliability model accurately, this paper uses hybrid Weibull model for reliability modeling. In order to estimate the parameters accurately, the initial values of model parameters are calculated by graphical Method. The search space of the genetic algorithm can be determined based on the initial values. Then the optimal model parameters can be iteratively solved by genetic algorithm, and the hybrid Weibull model can be obtained. Finally, by comparing with the single Weibull model, the hybrid Weibull model is more suitable for ATC reliability function. This paper provides a new idea for the reliability modeling of complex systems and provides a theoretical basis for the subsequent reliability evaluation.

Foodborne pathogens are the most common food safety hazards, and the traditional detection methods of foodborne pathogens are cumbersome, long waiting time and slow efficiency. This paper studies two common foodborne pathogenic bacteria, Brucella and Escherichia coli. LightGBM algorithm combined with Principal Component Analysis (PCA) was used to analyze Raman spectrum sample data to solve the problem of classification and detection of foodborne pathogens. The results show that LightGBM algorithm has excellent detection rate. Compared with traditional machine learning algorithm models such as Decision Tree, Random Forest and XGBoost, LightGBM algorithm has the following advantages of low memory consumption and model training speed fastly, and the model accuracy rate reaches 91.23%.

The detection of insulator defects in transmission lines presents greater difficulty and challenges compared to regular object detection tasks due to the small size of defects in the image and complex background. In this paper, we propose an insulator defect detection algorithm based on multi-scale feature fusion optimization to address the problem of extracting effective information from insulator defect images with complex backgrounds. The algorithm incorporates a multi-scale feature fusion optimization module into the CornerNet network, which enhances the network's ability to capture global contextual information and local detailed information of the insulator feature map. Experimental results demonstrate that the proposed algorithm achieves an average precision of 93.65% for insulator defect detection with an inference time of 32.15ms, providing high detection accuracy and inference speed. The proposed approach has certain reference value for insulator defect detection in transmission lines.

This study aims to enhance fault localization in distribution networks to ensure reliable electricity supply. Two fault location methods, range location, and section location, have been analyzed. The section location method, known for its fault tolerance, has been the focus of investigation. The Matrix algorithm and intelligent optimization algorithms have been explored to improve fault location accuracy. Three key contributions have been proposed: establishing a correlation model between low-voltage customer power outage data and fault section variables, developing a robust fault location model based on functional relationships, and conducting simulation experiments to demonstrate enhanced accuracy and robustness. The research outcomes provide practical insights into enhancing power system reliability.

Based on an analysis of the functional structure of the multi-way valve, this study determines the test items according to relevant test standards and requirements. It establishes a mathematical and simulation model to optimize the test process by shortening the test time for multi-way valves. The Gray Wolf algorithm is selected for optimization. To overcome limitations in the basic Gray Wolf algorithm, this research incorporates the good point set method, weighted distance updating, and variable neighbor searching algorithm to enhance the algorithm's performance. MATLAB simulations demonstrate the improved Gray Wolf algorithm's superiority.

At present, in the promotion and popularization of mobile applications, the application of user profiling technology is difficult and inefficient. In order to deepen the application of user profiling technology and reduce the difficulty of using user profiling technology, this paper proposes a hierarchical clustering user profiling construction model based on the improved K-means algorithm. The model focuses on constructing mobile application user features through hierarchical clustering algorithms to improve the computational speed and accuracy of user profiling model construction. Finally, the model is experimentally validated using a high-voltage power user sample set in the field of power marketing, The experimental results indicate that the hierarchical clustering user profile model can effectively improve the accuracy and computational speed of clustering and profile construction applications, making contributions to the promotion and application of user profile technology in power enterprises.

Liver cancer is one of the common diseases. It is of great significance to realize the automatic and accurate segmentation of liver and tumors in clinical medicine. The boundary blurring and the difficulty of feature extraction in CT images of liver and tumors are perplex, an improved medical segmentation model SKE-Unet++based on Unet++ was proposed. The SKE-Unet++ model adopts the connection mode of full-scale feature fusion and integrates coarse-grained semantics and fine-grained semantic information extraction features at full scale. In order to extract more important features after convolution, the SE (Squeeze Excitation) module is added to enhance the channel features of the input feature map. The foreground sample in CT image is much smaller than the background sample, cross entropy loss function and dice loss function are combined to solve the class imbalance problem. Compared with the SKE-Unet++ model, the Dice, Jaccard and ASSD evaluation indexes of liver segmentation task in Lits data set increased by 1.03%, 0.29% and 0.2897mm respectively, and the three indexes of liver tumor segmentation task increased by 7.54%, 9.33% and 1.2077mm respectively, which confirmed the validity of the model, and provided reference for automatic segmentation of liver cancer medical images.

Due to the low prediction accuracy of the existing Back Propagation Neural Network photovoltaic power prediction model, it cannot effectively reflect the intermittency and fluctuation of photovoltaic power range. The photovoltaic prediction system proposed in this paper utilizes the Pauta criterion to identify anomalies in historical input data and uses the genetic algorithm back propagation neural network algorithm for repair and prediction. The method was validated on a 100MW photovoltaic power plant, example analysis shows that genetic algorithm back propagation neural network can effectively improve the accuracy of photovoltaic daily power prediction compared with back propagation neural network.

Based on the classical Paillier homomorphic encryption algorithm and the ciphertext multiplication operation, it has the computing ability to deal with complex ciphertexts. It has also improved the efficiency of the algorithm based on ensuring support and confidence. By adding false data to the original plaintext data, the fuzzy degree of the original database is improved so that the ability of the model to deal with analysis attacks is effectively strengthened. MREclat algorithm is adopted to mine association rules, which improves the efficiency of system mining and increases the accuracy and reliability of the results. The experimental results have shown that although the proposed algorithm has obvious advantages in privacy protection, it can be used to protect user data in cloud computing.

Maze generation algorithms play a significant role in video games with mazes. Most of the existing methods focus on improving the performance of nonuniform maze creation algorithms. However, generating different mazes with equal probability through uniform algorithms is crucial to increase the user’s gaming experience. In this paper, we develop a hybrid of two uniform maze generation algorithms, Aldous-Broder and Wilson, by alternating between the two at some critical running phase to reduce the execution time. We tested several turning points and analyzed the results. The experimental data shows that the performance is better compared to either algorithm by itself when the percentage of visited cells reaches a specific value.

FP-Growth algorithm is a classic algorithm in the field of data mining. Compared with Apriori algorithm, all frequent item sets can be obtained by traversing the data set only twice, which improves the efficiency of data mining. However, in practical use, if there are many Frequent one item sets in the data set, the data set is scattered, and a lot of time will be wasted. Moreover, it will cause the problem that there are too many child nodes when FP-Tree is formed, and it is easy to overflow. In view of the above shortcomings, in this paper, a new FP-Growth algorithm based on Data Strong Association (SDA-FP-Growth algorithm) is proposed., which uses frequent two items set for operation integration and merging of the same leaf nodes, which is conducive to the subsequent frequent item set search. Experiments show that the SDA-FPGrowth algorithm reduces the running time and memory space.

This paper adopts a combined optimization approach using Simulated Annealing and Genetic Algorithm based on the travel purpose as a determining criterion. It selects one destination among multiple travel destinations and derives the optimal route based on the latitude and longitude information of the destination. The research aims to overcome the limitations of traditional navigation software, which requires explicit destination input to plan a route, and achieve intelligent route planning based on travel purposes. By identifying the travel purpose and selecting appropriate destinations to form the optimal solution, this study successfully addresses the limitations of conventional travel planning. The combined algorithm makes full use of the global search capability of Simulated Annealing and the optimization effectiveness of Genetic Algorithm, achieving superior travel routes by optimizing the initial solution. Experimental results and comparative analysis demonstrate that the travel route planning method based on the combination of Simulated Annealing and Genetic Algorithm exhibits significant advantages in solving complex travel problems, providing users with intelligent and efficient travel decision support. This research contributes to the innovation and improvement of travel planning technology, bringing greater convenience and comfort to the travel experience.

Aiming at the problems of weak ability to jump out of local optimization and lack of randomness in grasshopper optimization algorithms, this paper proposes a multi-strategy Enhanced Grasshopper Optimization Algorithm (EGOA) based on Brown random walk. The EGOA incorporates multiple strategies, including a Brownian random walk, a nonlinearly decreasing strategy with perturbation, and an improved optimal value guidance step. These innovations enhance the algorithm's ability to escape local optima, increase randomness, balance exploration and exploitation, and expand individual search ranges. Experimental results demonstrate that EGOA surpasses other swarm intelligence algorithms in benchmark function tests.

Currently, various energy sources such as electricity, heat, gas, and water in cities are relatively independently supplied, making it difficult to meet the demand for integrated energy utilization in new smart cities. The urban energy supply structure urgently needs to evolve from single transmission to gradual integration and coupling of multiple flows, which highlights the pressing need to enhance capabilities such as accurate perception of energy status data, trustworthy fusion and sharing, and precise energy efficiency services. These capabilities will greatly contribute to the improvement of overall urban energy efficiency and economic structure optimization. This article primarily focuses on the scheduling of computational resources between edge servers and terminal devices. Edge computing reduces data transmission latency and enhances computational agility during the computation offloading process. However, the computational capacity of edge servers is also limited, and in certain specific computation offloading scenarios (such as ultra-dense networks), interference may arise, leading to unexpected transmission delays. Therefore, it is not advisable to offload all tasks for execution on edge servers; some tasks should be handled by the terminal devices (SMD). Despite consuming more energy through local execution, this approach eliminates the need to consider data transmission time, thus significantly boosting task responsiveness. This article aims to design a task offloading scheme that is aware of terminal energy consumption and involves scheduling computational resources between edge servers and terminal devices.

Aiming at the problem of increased network delay due to congestion caused by link collision caused by improper traffic scheduling in the industrial Internet, this paper proposes a flow scheduling strategy based on probability-based data flow path planning within a period of time. On the basis of the software-defined network centralized control architecture, the controller collects the switches and edges’ node information in the topological network to meet the QoS of different services as much as possible while ensuring that more services are responded. In the scheduling period, the congestion probability of each link is introduced, and the relative delay is obtained by combining the delay required by the link. The optimal path under the current strategy is selected to optimize the result of data flow scheduling and reduce data flow transmission delay.

In order to solve the problem that computer software system often makes mistakes in disease coding, this paper puts forward an error analysis method of disease coding in computer software system. Firstly, a vocabulary database based on the third volume of ICD-10 is established, and the clinical diagnosis name and disease code checking rules are analyzed. Using the method of case analysis, this paper analyzes various factors leading to disease coding errors; Further induction and analysis, find the path of error correction. Accurate coding of diseases depends on the joint efforts of clinicians and medical staff in coding and standardized management, which has a great impact on diseases through accurate medical records.

Traditional rainfall-runoff prediction methods are mainly based on a single data source, such as only using meteorological observation to analyze and predict numerical data. However, this method is often affected by factors such as data missing and noise interference, which leads to low accuracy of prediction results. To solve this problem, paper proposes a rainfall-runoff prediction model based on multi-modal data fusion. The model uses image data as a new modal to forecast rainfall-runoff, and then uses Transformer and CNN to extract important information of digital data and image data. Since image data can obtain more information about vegetation cover area, in this model, vegetation features are used as new features to fuse with other feature information to improve the accuracy of prediction results. The experimental results show that proposed method achieves good results in single-day average runoff prediction in terms of performance and accuracy.

Existing Long-Tail classification (LT) methods ignore attribute class imbalance and focus only on addressing class imbalance where the head class has more samples than the tail class. In fact, even if the classes are balanced, the samples in each class can still be long-tailed due to attribute variability. The latter is fundamentally more general and difficult than the former, not only because attributes are implicitly included in most datasets, but also because they are more complex to construct. Therefore, we introduce a new research question, generalized long-tail classification (GLT), to jointly consider these two imbalances. And the ASG model is proposed to deal with the long-tail problem of attributes on the data set. ASG is a generalized class rebalancing method, which aims to balance the classes and attributes of the training data. It uses a feature displacement model to extract feature displacements, and then generates new samples according to the distribution of attributes in each class. Experiments prove that ASG has achieved good results on various datasets.

This paper describes a new biologically plausible parser based on the Assembly Calculus, a recently proposed computational framework implemented through the spiking of neurons. We demonstrate this parser can support Kleene star, an essential operator of regular expression. While the original parser proposed by Mitropolsky doesn’t support Kleene star. Hence our parser can deal with much more regular languages than the original parser. Our discussion suggests that the parser can be extended to parse some center-embedded sentences.

To study the effect of burst height on the performance of metal jet penetration into the moving target, finite element models of metal jet penetration into the moving target with different speeds at different burst heights are established using finite element simulation software, and numerical simulation is performed. The results show that with the increase of burst height, the metal jet is elongated, the velocity of the head and neck of the jet decreases and the diameter shrinks; when the target panel is stationary, the penetration depth of the metal jet into the target panel increases continuously with the increase of burst height; when the target panel is moving, the burst height for the maximum penetration depth of the metal jet corresponding to each moving speed of the target panel is not consistent with the burst height for the static maximum penetration depth, and with the increase of the moving speed of the target panel, the burst height for the maximum penetration depth of the metal jet decreases continuously; smaller burst height ensures the speed and diameter of the head and neck of the metal jet, enhances the stability of the metal jet penetration into the moving target and the resistance to interference, and improves the penetration performance of the metal jet into the moving target.

In the field of facial expression recognition, the subjectiveness of annotators leads to the phenomenon of label uncertainty in dataset samples. In this paper, the training Algorithm of facial expression Dataset based on variable weights is used to complete the label correction for uncertain samples. This algorithm can fully exploit the training potential of the dataset through the change of loss function weight in the training process. The algorithm in this paper achieves excellent results on the facial expression dataset RAF-DB.

The spectral radius of a matrix is widely used in numerical analysis, graph theory, stability theory, and other related fields, and is a rather active topic in matrix theory research. In this paper, we establish a smoothing algorithm to calculate the spectral radius of a non-detective nonnegative irreducible matrix by constructing a special matrix. The effectiveness of the algorithm is demonstrated by a numerical arithmetic example.

The recursive result of multi body space system dynamics is a dynamic process. In this process, multiple discrete points can be used to analyze the research objects. The space system dynamics of multiple objects is a discipline that studies the motion laws of multiple objects interacting in space. In the development of Space technology, the application of space system dynamics includes spacecraft design and orbit control. This article explored a recursive calculation method for space multi body system dynamics based on data analysis, with the aim of providing new computational ideas for the aerospace field. This article mainly analyzed and designed spatial multibody systems and compared them using recursive algorithms and Lagrange methods to obtain the advantages of recursive computation. The experimental results showed that the longest CPU (Central Processing Unit) time consumed by the recursive method was 9.3 seconds, and the shortest was 1.9 seconds.

Due to the excessive cost of data collection as well as annotation in dedicated domains, artificial intelligence model training is difficult to achieve optimality with insufficient data. To optimize this issue, a text generation data augmentation method based on the BERT model is proposed in this paper for augmenting the small amount of available annotated data. The optimization of this data augmentation method is demonstrated by experiments. In a text classification experiment, this data augmentation method can improve the training effect of the source data by 2.9%.

Uncertainty quantification is an essential method for sample-efficient deep reinforcement learning. There is a growing literature on uncertainty-based deep reinforcement learning algorithms, but many of the previous approaches failed to capture different sources of uncertainty. We highlight why this can be a crucial shortcoming for sample-efficient algorithms and provide a sophisticated analysis of the uncertainty in the interaction between agent and environment. Based on that, we propose Weighted Bootstrapped DQN, an exploration-efficient method that combines network ensembles and variance weighting. We use aleatoric uncertainty estimation together with epistemic uncertainty to improve the exploration ability of the algorithm. We prove that our new approach has a significant improvement in sample efficiency on different gym tasks, even compared with the previous state-of-the-art approaches.

With the rapid development of network technology, network-based applications are becoming more and more diversified, and network security issues are becoming more and more significant. It is an urgent task to study methods to protect system security. Log system, as an important part of information system, can record all behaviors generated by the system, and most of the security risks in the system can be found by analyzing log system. To achieve the purpose of improving the system security performance. As an active network security protection measure, anomaly detection technology can detect abnormal behaviors and respond in time to effectively prevent abnormal behaviors and reduce the risk of system destruction. In this paper, public offline logs are used as data sources, common anomaly detection algorithms are studied, and effective information is obtained from massive log data to perceive the situation of network security.

In order to reduce the average delay and link utilization of data scheduling and improve the throughput, an intelligent scheduling algorithm for heterogeneous sensor communication big data based on LSTM and cloud computing is proposed. In the cloud computing environment, the amount of scheduling tasks is calculated in partitions, and based on the calculation results, the big data transmission load of heterogeneous sensor communication is predicted by using the Long Short-Term Memory network (LSTM). Get the working status of sensor nodes, classify all nodes, and form a cluster from sensor nodes in the same working status. Divide the heterogeneous sensor communication network into different regions according to different clusters. Calculate the energy consumption of nodes when they are cluster heads and non-cluster heads in a scheduling and the total energy consumption of heterogeneous data scheduling, so as to achieve intelligent scheduling of big data of heterogeneous sensor communication. The experimental results show that the proposed method has low average latency and link utilization, and high throughput, which verifies its application effectiveness.

With the massive use of mobile IoT technology and shared thinking innovative electricity consumption mode, shared electricity consumption smart devices have come into being, and the identification of abnormal electrical energy metering data in the process of shared electricity consumption collection has become an urgent research problem. In this paper, by studying the method of identifying abnormal electrical energy metering data in shared electricity information collection system, we select the improved particle swarm algorithm to optimize the parameters of support vector machine kernel function, construct the power quality disturbance model, and implement the classification of abnormal electrical energy metering data collected by electricity information system; use LOF algorithm to calculate the abnormality factor, and use the disturbance model determined by fly away abnormal intelligent analysis method to judge whether the displayed value of electrical energy meter The LOF algorithm is used to calculate the abnormality factor, and the disturbance model determined by the fly-away abnormality analysis method is used to determine whether the displayed value of the energy meter is abnormal or not. The experimental results show that the method of this paper is more accurate in classifying abnormal data and can better prevent the occurrence of error judgment, which can effectively improve the quality and efficiency of abnormal judgment of electric energy metering data and better realize the accurate measurement of shared electricity.

In order to improve the accuracy of traffic flow prediction on dangerous curve sections and shorten the time consumption, this paper designs a traffic flow prediction algorithm on dangerous curve sections based on Random Forest algorithm and open-source cloud computing platform. Firstly, collect traffic flow data for dangerous bend slope sections, and design a traffic flow data storage system architecture for dangerous bend slope sections based on open-source cloud computing using Ceph open-source distributed storage system; Then, the stored data is repeatedly cleaned, and the cleaned data is used as the test data. Finally, a traffic flow prediction algorithm based on Random Forest is designed for dangerous curve sections. So far, the design of traffic flow prediction algorithm for dangerous curve sections based on Random Forest algorithm and open-source cloud computing platform has been completed. The experimental results show that the overall effect of the traffic flow forecast of the dangerous curve slope section designed in this paper is good. When the number of vehicles reaches 150, the prediction error of the proposed method is only 6.17%, and the prediction time is 12.4 s. It is proved that the design method in this paper can significantly improve the traffic flow prediction of dangerous curve slope sections and can be used in dangerous curve slope sections.

The continuous development of computing and communication technologies, as well as the emergence of new applications such as autonomous driving, augmented reality, and industrial Internet of Things, have posed significant challenges to the computing and storage capabilities of terminal devices. We need a new computing paradigm to provide high-speed and low-latency computing services to meet the needs of these new applications. While cloud computing technology offers abundant computing power, it often fails to meet the latency requirements of these applications due to the long distance between cloud servers and terminals. To address this issue, the network paradigm of edge computing has been introduced. One of the key problems in edge computing applications is how to offload tasks generated by terminals effectively. Previous research has shown that reinforcement learning methods are effective approaches to tackle computation offloading. In this article, we provide a comprehensive survey and summary of the application of reinforcement learning/deep reinforcement learning in the context of task offloading and analyze possible future directions.

Traditional particle swarm optimization has attracted attention in various fields because of its relatively simple form and flexible parameter setting, but it also has the disadvantages of slow convergence speed and easy to fall into local optimization in the face of large-scale multivariate data. To solve this kind of problem, a chaotic multi-group optimization algorithm (CM-PSO) based on Graphics Processing Unit (GPU) is proposed. In the algorithm initialization stage, chaotic mapping is introduced to enhance population diversity, and then the population is divided into multiple small subgroups according to the idea of island model, and the Feng's topology is adopted within each subgroup to improve the search efficiency and reduce the computational complexity. Finally, the CUDA stream (streams) technology is used to realize grid-level parallelism, further improve the degree of algorithm parallelism, and improve the algorithm performance while ensuring the accuracy of the algorithm.

Cloud detection of remote sensing images is an important preprocessing step for remote sensing image applications. While machine learning (deep learning) methods improve the accuracy of cloud detection in remote sensing images, they put forward higher requirements for the quantity and quality of data sets. However, if you want to reconstruct the dataset by yourself, you generally need to select the training samples by manual visual interpretation, which will consume a lot of manpower and material resources. Previous research mainly focused on the optimization of cloud detection algorithms, and few studied the extraction of training samples. In order to construct a cloud detection dataset, this paper proposes an Erosion and Diversion-based semi-Supervised Learning (EDSL) model. Fmask algorithm is used to automatically obtain cloud masks from Landset-8 remote sensing images to construct a cloud detection dataset. The results show that the dataset constructed by EDSL model is better than Fmask algorithm. The precision of the dataset constructed by EDSL model reaches 0.93, which exceeds Fmask. At the same time, we use SVM algorithm to conduct experiments on this data set, and the results are equivalent to the data set constructed by human vision, which proves the scientific nature of this method.

Ellipse fitting is widely used in image processing, machine vision and other fields. Aiming at the problems of large data computation, high complexity and longtime consumption in the implementation of ellipse fitting algorithm by programmable gate array (FPGA), this paper proposes an FPGA implementation method of ellipse fitting based on least squares. This method is based on the principle of least squares. In the ellipse fitting problem under constraints, the generalized eigenvalue and eigenvector problem of the sixth-order matrix is avoided, and it is transformed into the eigenvalue and eigenvector of the general third-order matrix. Householder transformation and Givens rotation are used to greatly reduce the computational complexity. The simulation results show that the method still has high accuracy when using fixed-point number calculation.

Artificial light sources have intensified light pollution, diminishing visible stars and threatening biodiversity and ecological balance. It is crucial to establish an evaluation index of light pollution and put forward effective treatment strategies. This study establishes an evaluation model for light pollution, considering extensive data and principal component analysis, and the clustering algorithm is used for analysis. Then we came up with three solutions to the light pollution issue and spoke about how each might affect its prevention and management. In addition, we applied these three strategies in two different regions and came up with the most effective intervention strategies. Finally, two typical regions are selected to illustrate the pathways through which different strategies affect the final results.

The New York Times presents Wordle, a word game where players guess a 5-letter word in six attempts, receiving feedback after each guess. This paper presents a comprehensive study using an ARIMA time series prediction model to forecast the number of reports in the Wordle game, leveraging autocorrelation, lag, and averaging of data to make accurate predictions. Additionally, four-word attributes potentially influencing report numbers are extracted and correlated. OLS regression models and Pearson correlation coefficients are employed to analyze the impact of these attributes, highlighting the frequency of solution words and the number of vowels as significant factors in predicting the results, while other effects were found to be negligible. The research findings are validated through statistical tests, offering valuable insights into Wordle game dynamics.

The imbalanced food dataset is one of the various imbalanced datasets that are commonly found in the real world. For imbalanced datasets, we usually want algorithms to identify minority class samples in the imbalanced datasets, but traditional classification algorithms often suffer from performance degradation, overfitting and other problems. To address these problems, we propose a method based on Gaussian distribution resampling combined with a machine learning classifier. The method generates new samples in the form of Gaussian distribution by metering minority class samples in the unbalanced dataset, and after selecting the appropriate anchor samples, a balanced dataset is obtained. Finally, the balanced dataset is classified by a classifier. By conducting experiments on two real imbalanced food datasets, our method outperformed the classification performance of mainstream resampling methods on both datasets, achieving the best AUC and G-mean values. The experimental results demonstrate that our method can effectively identify minority class samples in the imbalanced food dataset and has some application value in the food field.

Computing power network is an emerging paradigm to meet the increasing computation demands and improve the system performance. Video stream processing is one of the important scenarios of computing power network, which puts strict demands on the limited edge computing and network resources. In this paper, a decision-making agent is constructed by a deep reinforcement learning algorithm, and the offloading location and resource allocation of video processing tasks are dynamically decided based on the system load, task status and service quality. A unique advantage is that the algorithm is model-free and does not rely on numerous system assumptions. By constructing the test-bed system, it is verified that the proposed method can achieve the optimization of task delay and energy consumption in real AI video stream processing scenarios. Future research directions are discussed in terms of the theory and system design.

We propose an improved AlexNet network model, to address the problems of low denoising performance of traditional LeNet-5 neural networks in removing random noise from seismic data. The network retains the original eight-layer calculation depth and uses ReLU as the Activation function to reduce the convolution core and the number of nodes in the convolution layer, thus obtaining higher noise feature extraction accuracy. The network trains the network with 10000 seismic data, tests the network with 1000 data, and optimizes the network. Experiments were conducted using a wide range of Marousi2 seismic data, and the results showed that the proposed network has good denoising performance. Compared with traditional wavelet algorithms, SVD, and LeNet-5 networks, experimental results show that the proposed network can achieve higher PSNR and SNR values and has better seismic data denoising performance compared to the above networks.

ARM architecture has shown potential performance in HPC applications, as demonstrated by the Fugaku supercomputer's ranking in the TOP500 in 2020. This paper explored the performance advantages of the Kunpeng 920 cluster made by the Huawei company. STREAM, HPL, and HPCG benchmarks are chosen for our evaluation. Additionally, this paper used novel HPC packages such as the BiSheng compiler, Hyper MPI communication library, and Kunpeng KML math library developed by Huawei. And they are adopted by 10 HPC applications from the Molecular Dynamics, Meteorology, and Life Sciences fields, including GROMACS, CP2K, LAMMPS, AmberTools, WRF, NEMO, CESM, FALCON, BWA, and PASA. The HPC applications are used to evaluate the computing performance and scalability of the Kunpeng 920 cluster. And this paper compared the results with those of the x86 cluster consisting of Intel Xeon Gold 6248. The results show that the Kunpeng 920 cluster has 27.9% higher memory bandwidth and better computing efficiency than the x86 cluster in benchmarks. Most of the HPC applications on the Kunpeng 920 cluster have better computing performance than the x86 cluster, to varying degrees.

In a distributed digital identity system, there are various types of credential storage entities. With the development and popularization of distributed digital identities, data silos have emerged among multiple credential storage entities. To break this dilemma, this paper designs a distributed digital identity private sharing method based on an efficient Random 1-out-of-n OT extension protocol. The method uses the data obfuscation algorithm constructed by Permutation-based Hashing to design a two-party private data sharing protocol, which outputs the intersection of private data sets to the two parties. After experimental analysis, on a scale of ten million data, this method can better meet the needs of the real world.

In order to solve the problem that providers and consumers both meet their requirement in a maximum way, we need to match the best providers and consumers to each other, so as to achieve the maximum needs. For this problem, this paper proposes an intelligent distribution technology of edge computing task based on the competitive mechanism. Firstly, it models the problem, constructs the objective function and the constraint conditions, and uses the simulated annealing algorithm to realize the global optimization combination solution. The experiments show that this method can provide a better task assignment method for providers and consumers.

In credit card, shopping, and insurance systems, a small amount of fraud may sometimes exist in the massive transaction data, so fraud detection is widely used in these systems. Extreme Gradient Boost (XGBoost) is outstanding in the competition, many scholars use it in various different fields of classification and prediction. The hyperparameters set by XGBoost determine its performance strength or weakness, and the many hyperparameters make it difficult to have good choices. In this paper, we use the improved Lévy sailfish optimizer (LSFO) to solve the parameter selection problem. As a relatively new intelligent optimization algorithm, the sailfish optimizer chooses spiral updating to optimize the search process of the sailfish and adds Lévy flights to increase its performance. To validate the improved performance, several common optimization algorithms such as PSO, HHO, and WOA are compared using the Lévy Sailfish Optimizer. A fraudulent dataset is used to evaluate the performance of Lévy Sailfish Optimizer Optimization XGBoost (LSFO-XGBoost) in comparison with Decision Tree (DT), Random Forest (RF), SFO-XGBoost and PSO-XGBoost. In LSFO-XGBoost hyperparameter optimization, due to its long training time, the distributed Spark framework is applied to improve its training in a distributed manner. The experimental results show that the proposed LSFO is effective in solving the XGBoost hyperparameter problem, and the distributed LSFO-XGBoost algorithm is better in terms of time consumption and performance.

With the development of the economy, the vacancy rate of housing in underdeveloped areas of China exhibits characteristics of uneven distribution and seasonal variation, which poses difficulties in accurately assessing the health status of the real estate market. The widespread use of smart meters has made it possible to collect accurate electricity consumption data. In this paper, based on a large amount of electricity consumption data, we propose a discrimination and prediction model for the vacancy rate of houses in Guangxi Province and utilize this model to predict the vacancy rate, providing reliable evidence for evaluating the health status of the real estate market.

Edge computing, as a complement to cloud computing, has many unique advantages. In Mobile Edge Computing (MEC) scenarios, tasks on terminal devices can be offloaded to edge servers for computing purposes, reducing the shortage of computing resources on terminal devices. Moreover, the resulting delay is much shorter than uploading the task to the cloud center directly. However, many scenarios in reality show that there are dependencies between tasks, and a task cannot be calculated until all its predecessor tasks have been calculated. In this paper, we summarize a variety of task dependencies. At the same time, we build an efficient task offloading model based on two levels of edge servers with different functions. In particular, the second-level edge server has the routing decision function, which provides us with diversified offloading methods. Finally, we used improved NSGA-Ⅱ for task offloading. Compared with other offloading methods, our algorithm performs better in latency and energy consumption.

Carbon emissions from transportation are an important part of urban carbon emissions. Vehicles, as the most active part of urban traffic, are important for refined estimation. Refined estimation can help provide an opportunity for more specific emission reduction strategies, but existing methods are unable to achieve this due to the lack of use of multi-source data. In this regard, this paper proposes a refined estimation method based on multi-source data fusion around the vehicles. First, consider the latent impact of electric vehicles on carbon emissions during their operation. Then, an improved Data Aggregation Model is proposed to represent the core part of carbon emissions. Subsequently, a carbon emission estimation model based on BP neural network is established, which uses the multi-source data of vehicle-network-road-facilities to obtain the final output. Finally, an innovative refined estimation method is proposed to enhance the results generated by the carbon emission estimation model. With results testing with real data, the accuracy of the estimation model and the refined estimation method is verified by confidence interval method.

Depression is a common psychological disorder. In order to detect and identify the tendency of depression in text as early and accurately as possible, a method for depression tendency detection based on multi-feature fusion of BERT word embeddings is proposed. An emotion classification model is employed for vector representation of depressive text, extracting vectors that fuse multi-level semantic features and emotional features. The TextCNN and BiLSTM-Attention modules are established to extract local and contextual features of depressive text. Finally, multiple features are fused to determine the tendency of depression in the text. Experimental results demonstrate that the proposed method, which fuses multiple features of depressive text, achieves an improvement of approximately 5% in various evaluation metrics compared to traditional deep learning methods.

In recent years, the proliferation of smartphones and advancements in internet technology have led to the emergence of various internet applications, resulting in a surge in data volume. However, this expansion has also brought about information overload. To address this issue, recommendation systems have gained prominence, leveraging recommendation algorithms to extract user-relevant items from extensive datasets based on user interests. These systems have found applications in various domains such as e-commerce and social networks. However, most collaborative filtering algorithms are tailored for sparse user-item rating matrices, neglecting scenarios with densely rated popular items by heavy users. This paper presents an improved collaborative filtering algorithm for dense data scenarios. It utilizes the K-Nearest Neighbors (KNN) approach, focusing on user rating predictions to tackle challenges arising from dense data environments. The proposed algorithm divides user ratings into liked and disliked item sets, enhancing prediction accuracy by calculating similarities separately. Experimental evaluations are performed using a real-world dataset from the "Kuaishou" video sharing app, demonstrating the algorithm's superiority in terms of F-Score compared to traditional collaborative filtering methods. The algorithm's simplicity, scalability, and performance are highlighted, showcasing its effectiveness in addressing recommendation challenges in real-world settings. Future research directions could explore its applicability across diverse datasets and domains.

In order to improve the quality of decision-making on the protection of aerospace materials in land aviation units, the consumption of training aerospace materials in army aviation units is predicted and analyzed. With an army aviation unit training aerospace material consumption as the main research object, the influence factors of helicopter aerospace material consumption are analyzed, partial least squares regression and Markov chain are used to establish a prediction model of helicopter aerospace material consumption, and through the historical data of aerospace material consumption in helicopter training, aerospace material consumption in helicopter training is predicted, and the results show that the partial least squares regression and Markov chain model, which can predict the consumption of helicopter aerial materials more accurately.

In the burgeoning era of deep learning and computer vision, image classification has become paramount across diverse domains. The CIFAR-10 dataset, emblematic in image classification, garners widespread attention from scholars. Yet, the imperative of data preprocessing, especially sorting, cannot be understated, especially for expansive datasets like CIFAR-10. This paper sheds light on a novel optimization technique addressing the snags encountered by traditional Python sorting algorithms with CIFAR-10. Rigorous experimental validations underscore remarkable enhancements in sorting efficacy and precision, paving the way for robust preprocessing in ensuing image classification tasks.

This paper investigates a layout planning method for virtual maintenance environments based on the computer ant colony algorithm, incorporating the regularities of human perception psychological activities. Through in-depth research on the ant colony algorithm and human perceptual psychology, we propose a layout planning method based on the computer ant colony algorithm to enhance the effectiveness and quality of layout planning. We put forward two key observations: firstly, the principles of human perceptual psychology play a crucial role in the visualizable and comprehensibility of layouts; secondly, the ant colony algorithm can simulate the cooperative and rational reasoning processes in human perceptual psychology, thereby optimizing the decision-making process in layout planning. To validate our method, we analyze the impact of spatial perceptual experience and sensory systems and establish a two-dimensional layout mathematical model. The ant colony algorithm is used for optimizing layout solutions. Through case analysis and experimental verification, the effectiveness of this method is demonstrated. The proposed method can improve work efficiency in the layout planning of virtual maintenance environments and provide a good spatial experience for maintenance work. Furthermore, future research directions are proposed, such as further exploring the regularities of other perception psychological activities to improve the quality and effectiveness of layout planning. This study has guiding significance for promoting the development of virtual maintenance environments.

In view of the current problems of untimely information updates, incomplete interactive responses and impractical classification systems often arising from the question-and-answer sections of government websites, this paper provides a research direction for the organic integration of online question-and-answer and text classification by taking natural language processing technology as the research perspective and combining the construction and application of deep learning algorithm models. This paper improves the algorithms in text representation and deep learning models from the perspective of text representation and feature extraction. In response to the shortcomings of Bert model which weakens the text location information, LSTM model can effectively learn the dependencies on the observation sequence and extract the global features of the preceding and following texts. Therefore, this paper combines the Bert and LSTM models for feature extraction and introduces the Attention mechanism to filter and fuse the extracted features in depth to build the Bert-LSTM model. The classification results of this model and the Bert model on the same dataset are compared, and it is proved that the new hybrid depth model can improve the classification accuracy of the questioned text.

Global warming, which is deeply related with our life, is becoming more serious overtime. Few scholars have used deep learning models to predict changes in temperature. Most of the existing studies are single temperature predictions and no mechanistic analysis has been carried out. Therefore, we have designed two models to predict the future temperature: ARIMA model and LSTM model. With these two models, we have provided a scientific prediction of future temperature increase. In addition, we also performed a correlation analysis to look for the factors influencing temperature. Considering all models together, we agree that global temperatures will reach 20°C by 2050, and we find that anthropogenic activities have a greater impact on global warming.

Oracle bone inscriptions are the oldest systematically written characters in China, and their periodization and chronology are essential tasks in historical research. However, current deep-learning models face significant challenges when they are used to recognize oracle bone images directly. These models can only achieve an accuracy of no more than 45%. As the research on the periodization of oracle bones mainly relies on the characters inscribed on the bones, a domain knowledge-based oracle bone classification model is proposed that improves the recognition accuracy of Vision Transformer (ViT) by 5.28%. The proposed approach incorporates results obtained from oracle bone character detection to provide a more effective recognition object. Moreover, a novel model with a style network and category embedding is proposed in this paper that focuses on characters, resulting in a notable improvement in recognition accuracy of 16.01%. Our study reveals the importance of domain knowledge in the classification of historical documents, and it presents new insights into using artificial intelligence for assisting in historical document organization tasks.

The impact of age-related brain atrophy on the decline of cognitive function in older adults has been widely recognized. To investigate this issue, we conducted a study using machine learning algorithms to analyze Magnetic Resonance Imaging (MRI) data collected from a large UK Biobank sample of 6, 000 individuals, ranging in age from 48 to 84 years. Our methodology consisted of splitting the entire cohort into two groups based on brain age gap estimation from three image modalities: the Advanced Brain Age Group (ABAG) and the Normal and Healthy Brain Aging Group (NHBAG). Subsequently, using a mixture of experts to provide more clarity on the group heterogeneity of NHBAG, we identified four subtypes. These included the temporal lobe preservation subtype, the minimal atrophy subtype, the diffusion atrophy subtype, and the frontal lobe preservation subtype. These results emphasize the importance of exploring the complexity of age-related brain changes and their subtypes through advanced data-driven techniques.

In recent years, social network public opinion analysis technology, represented by text mining and sentiment analysis, has become an important tool to quickly understand public opinion. In this paper, we focus on the public opinion evolution in Sina Weibo and take the illustration event of textbooks in China’s primary schools in 2022 as an example. We crawl 13, 833 microblog hot comments on the illustration event of the textbook. Firstly, we pre-process the data, divide the evolution process into the outbreak period, the response period, and the final period, and give the word cloud and econometric analysis to display the data. Then, in order to study the public opinion evolution, we adopt the Latent Dirichlet Allocation (i.e., LDA) model and the lexicon-based sentiment analysis method. In particular, for each period, we capture the topic distribution of comments through the LDA model and learn the sentiment polarity through the lexicon-based sentiment analysis method. Finally, we utilize a knowledge graph to visualize results. The results show that during the outbreak period and response period, public sentiment is negative, and hot discussion topics are problems existing in the textbook, and also show their impact on education, and national ideology. In the final period, the discussion focuses on supporting the Ministry of Education to intervene in the strict inspection of elementary textbooks, and the public sentiment is positive.

Nowadays, cloud-rendering systems have attracted the intense interest in the public and been used in many areas including video games and online education. In this paper, we will focus on the streaming compression (i.e., command-and video-streaming), an important metric playing an indispensable role in the performance of Cloud-rendering Systems, especially for lowering the network bandwidth consumption. Based on the analysis of the characteristics of rendering commands and generated frames, we propose an effective method to compress them and reduce the bandwidth consumption. Finally, we will show the compression ratio we achieved when applying these compression methods.

Offline Reinforcement Learning (RL) aims to learn an optimal policy from a fixed dataset previously collected. Unlike in the online training process, the errors in value estimation from out-of-distribution actions (OOD actions) could not be corrected by interacting with the environment, which makes offline RL difficult to train. Prior policy constraint methods mitigate the errors by minimizing the deviation from the behavior policy, which in fact makes a trade-off between RL and imitation learning. However, while using the constraint term to avoid choosing OOD actions, the learned policy may be enslaved to the unfavorable actions from the dataset. In this paper, we propose a simple solution to this problem. Our method uses an imitation learning term introduced by TD3PlusBc as the constraint and re-weights it with a function of actions' advantage value to mitigate the influence of the unfavorable actions. To make the learning process stable, we also decouple the policy evaluation and policy improvement by using implicit q-learning, which modifies the loss function in a SARSA-style TD backup. Our method (ABAI) is easy to implement, fast to train, and computationally efficient. ABAI achieves state-of-the-art performance on the D4RL dataset, a standard benchmark for offline reinforcement learning, and shows a high ability to learn robustly from various datasets.

The Delay Tolerant Networks protocol (DTN) provides a solution for the integration of space and ground, where the transport layer uses the bundle protocol to connect different restricted networks and uses the storage-hosting-forwarding method to achieve reliable transmission of data packets. This paper uses Specification Description Language (SDL) formal description language to map the information delivery between different modules to transfer of states under different input conditions between state machines. Based on this, SDL simulation language is used to model and simulate the target. The correctness of the execution process of each module under given input parameters is dynamically verified through message sequence diagrams. Finally, the simulation verification results are analyzed and explained.

As the scale of distribution networks continually expands, there is an escalating need for the operational modality of distribution network operation ticket systems to transition from a traditional "manual experience-based scheduling" to a more advanced "intelligent scheduling" system. Many regions still rely on conventional methods for ordering, returning, and reporting, which lack robust network platform support. This limits work efficiency, puts additional strain on personnel and operation teams, and ultimately heightens their workloads. In response to these challenges, this study proposes the design of a network order process management system underpinned by mobile terminal technology. This system integrates voice input capabilities, target similarity algorithms, and the ability to switch between online and offline operational modes to construct a comprehensive distribution network operation ticket management system. The design includes both a network order management platform and a mobile operation platform for handling operation tickets. In terms of specific technology adoption, the project employs intelligent voice input combined with target similarity algorithms for voice-to-text conversion and command repetition tasks. This technological integration offers a convenient method for voice-to-text conversion, which will likely gain increasing favor in the digital transformation of operations. Furthermore, the project utilizes an online-offline dual-mode switching technology, serving as a potential reference for other mobile operation scenarios. This includes integrated data synchronization, equipment monitoring, and Short Message Service (SMS) alert technology to achieve real-time monitoring of offline operations. This operational mode has broad applicability and can be extended to other operational scenarios, such as mobile maintenance operations.

This article introduces a solitary adjustable parameter-based filled function technique tailored for constrained global optimization. The devised filled function strives to encapsulate the idiosyncrasies inherent in the optimization quandary, and the single parameter can be easily tuned during the optimization process. The paper explores the analytical characters of the filled function and presents an algorithm based on the filled function approach. Finally, numerical tests are implemented to demonstrate the effectiveness of the proposed filled function method.

Power customer complaint texts usually contain a large number of natural language texts, which may have problems such as polysemy of vocabulary, complexity of grammatical structure and subjectivity, and are difficult to classify. Therefore, a classification method for power customer complaint texts based on natural language processing technology and drosophila algorithm is proposed. The text is obtained through web crawler technology, the text is segmented by natural language processing technology, and the modal particles and stop word in the text are removed. The bag-of-words model is used to characterize the text and obtain the result of text feature representation. Based on the text preprocessing results, vectorize the complaint text of power customers. Finally, based on the text preprocessing and vectorization, the drosophila algorithm is used to realize the text classification of power customer complaints. The experimental results show that the proposed method has high text classification accuracy, recall rate, and F1 value, indicating that its classification effect is relatively accurate.

The current availability of images for the recognition of Chinese herbal pieces is very limited, making it challenging to meet the demands of the development of an automatic recognition system for Chinese herbal pieces, particularly one that uses artificial intelligence. This paper uses mobile device shooting and crawling technology to collect the image data of Chinese herbal pieces, uses multi-scale detail boosting algorithm to enhance the image, and finally uses rotation, mirror image and translation to expand the image data, and a dataset containing 64,000 images of commonly used Chinese herbal pieces is constructed, which is divided into 50 categories, which can meet the realization of artificial intelligence technology. In this paper, six convolutional neural network image classification algorithms are evaluated based on the constructed data, namely AlexNet, VGG, GoogLeNet, ResNet, ResNext and MobileNet. Experiments show that the classification accuracy of this dataset can reach more than 90% in six image classification algorithms, so the constructed dataset can be used to verify the effectiveness of the image classification algorithm of Chinese herbal pieces.

QR code is widely used in production management. For example, we scan the QR code printed on the fertilizer bag during the production of fertilizer to record production information. We propose a method to intelligently obtain the information carried by QR code through machine vision, combining deep learning and image processing technology. We obtain the original image through the industrial camera, locate the QR code area in the image based on the SSD target detection model, use the image processing method to rectify the QR code area, and finally decode it through the Zxing library. Through the real production environment test, the recognition rate of the method is 98.6%, and the average detection and recognition time of a frame image is 94 ms.

Mechanical manufacturing process knowledge exhibits complexity, detail-oriented nature, experiential characteristics, high learning cost, and continuous evolution, posing challenges in its management, utilization, and transmission. Enterprises currently face low knowledge reuse rates and high knowledge management costs. To address these issues, we propose a novel approach utilizing deep learning models to extract entities and relationships from unstructured knowledge, constructing a process knowledge graph. With deep learning's powerful comprehension abilities, we automatically extract critical information, organizing it into a structured knowledge graph consisting of 12 entities and 12 relationships. The process knowledge graph enables convenient knowledge transmission and sharing, increasing reuse rates and reducing redundant learning. It serves as an effective tool for process formulation, enhancing efficiency and enterprise benefits. Additionally, it facilitates prompt identification of suitable process parameters, accurate prediction of manufacturing issues, and timely adjustments. Moreover, the process knowledge graph fosters innovation by identifying optimization solutions and improving product quality and cost reduction. This approach offers significant competitive advantages and commercial value to the manufacturing industry.

How to provide individualized instruction and personalized teaching for students and effectively track their knowledge status in a targeted manner is a key concern in the field of education. The target of Knowledge Tracing (KT) is to use the learner's behavior in historical learning activities to model the students' educational process and forecast the future performance of students. Recently, self-attention has been used several times in the KT field. Traditional models have the problem of losing information under long sequence data. Inspired by Transformer, in this paper, we present a deep knowledge tracing model (MA-DKT) based on the multi-head self-attention mechanism, which employs the LSTM model combined with the multi-head self-attention mechanism to implement a deep knowledge tracing model that is able to catch the data characteristics of the students at different temporal scales. Our model works well on publicly available datasets and achieves some improvement in AUC.

Approaches and models of intelligent control have been developed to improve quality of daily life. Smart home solution has been designed to provide convenience for residents by recognizing their daily activities. In order to improve the performance of daily activity recognition, here we proposed an improved BiLSTM-Attention based daily activity recognition method which introduced dimension upgrading. Firstly, we used the feature extraction algorithm to obtain the relevant feature vectors. Secondly the feature vectors were transferred to the Bi-directional Long Short-Term Memory (BiLSTM) model for learning the time-series relationship between features and daily activity recognition. Thirdly, the BiLSTM model and attention mechanism were combined to improve the discrimination capability of daily activity features. Finally, the input data of the BiLSTM-Attention model was upgraded to improve the performance of daily activity recognition. The the F1-score of BiLSTM-Attention model was improved over the benchmark method by 5% using the public dataset CASAS_Cairo, and by 19% after the data dimension upgrading. On the public dataset CASAS_Aruba, the F1-score was improved 6% by using BiLSTM-Attention model and improved 11% by using additionally data dimension upgrading.

As a clean chemical energy source, LNG is widely used in daily production and life, so the study of LNG terminal and its risks is extremely important. In this paper, the fuzzy analysis method is used to establish a fuzzy analysis model of the potential risk of LNG terminal, by analyzing the weight of the potential risk factors of LNG terminals, it is found that natural disasters, unloading operation and power station environment are the most important factors causing the potential risks of LNG terminals, so that a series of potential risk control measures are proposed according to the risk factors.

In order to better improve the service quality of scenic spots and the travel experience of users, this paper uses sentiment analysis and text theme features to construct scenic spot portraits and user portraits and establishes the coupling relationship between scenic spot portraits and user portraits. The application of scenic spot portraits and user portraits can help tourism companies better understand the characteristics and needs of users and destinations, improve the personalization and accuracy of tourism products and services, and enhance the market competitiveness and development potential of enterprises.

This article presents the construction of a large-scale, high-quality Chinese Business Dataset (CBD) for knowledge graph construction in the Chinese business field. We consider the characteristics of the Chinese language and verify the effectiveness and reliability of the dataset through a series of experiments. The article discusses the process of extracting issues, entity links, and manual evaluation of the dataset. The authors conducted experiments using different relationship extraction algorithms based on the published dataset for training. The results suggest that Context-Aware is highly effective for relation classification in this task, while Bi-LSTM Attention may be insufficient. The use of knowledge graphs has enabled the organization of large amounts of structured and unstructured data, providing a more comprehensive understanding of business entities and their relationships. However, there are still challenges that need to be addressed in this field. Nevertheless, with the continuous advancement of technology and the increasing availability of data, further progress can be expected in this field.

At present, the on-site safety monitoring of power is mainly monitored by personnel through the whole process of surveillance video, but the use of manual detection method is not only a waste of time, but also prone to missing the situation, so that the personal safety of staff cannot be guaranteed. In order to realize intelligent recognition of workers’ behavior on the job site, an OpenPose based hazard behavior recognition method for power workers is proposed. The method extracts the key bone information of workers from video stream images, uses deep neural network to realize the human behavior and posture perception of workers in multi-person scenarios, detects and recognizes the violations of construction workers in real time, and issues warnings. The experimental results show that the proposed method can realize the accurate and real-time safety monitoring of the power field operators' behavior, ensure the personal safety of the field operators and the smooth progress of the power operation, and has certain robustness and generalization ability.

The digitization of emergency plans is an important foundation for improving the information level of emergency management in China. In response to the problems of low level of intelligence and reliance on traditional document management methods in most emergency plans in the water conservancy industry, this paper focuses on the technical path and methods of constructing a knowledge graph for flood control and emergency response plans. Based on this and combined with the construction of the knowledge platform of the digital twin project of the Dateng Gorge Reservoir, digital practices of flood control and emergency response plans for Dateng Gorge Reservoir were carried out. The application results provide intelligent support for emergency decision-makers in the smart water conservancy industry and effectively improve the scientific and accurate decision-making in water conservancy emergencies.

In the context of big data of power operation and maintenance, the labeling of operation and maintenance images reduces the difficulty of daily work of operation and maintenance dispatching staff and is widely used in various business platforms of the power grid. This paper studies an operation and maintenance image tag recognition technology based on deep learning. Combined with Convolutional neural network, it completes the screening and analysis of effective tags, helps the operation and maintenance control staff understand the analysis results of big data, can more efficiently complete the operation and maintenance data processing, and improve the work efficiency of the operation and maintenance staff and the stability of power grid operation.

In order to reduce the development and promotion cost of developers and accelerate the development speed of developers, emotion recognition technology is widely used in Internet and other scenarios, and an emotion recognition open platform based on deep learning is proposed. This article provides a detailed introduction to the design and implementation of an open emotion recognition platform. Overall, the platform has the following characteristics: it uses the Struts2+Spring+JPA framework, adopts interface-oriented programming, and has good scalability; The Convolutional neural network in the deep learning technology is used to realize expression recognition, which has good recognition accuracy; OpenID+OAuth authentication and authorization scheme is adopted to make the platform have better security. The results show that the ck+ dataset, which is different from the training dataset, is used to test, and the contempt emotion in ck+ is removed. The last three faces in each directory are regarded as the same emotion, corresponding to the emotion file, and the first face in each directory is regarded as the natural emotion. There are seven emotional types in total: anger, jealousy, fear, happiness, sadness, surprise and nature. After that, the face of each picture is extracted, and then the platform interface is used to identify it. Compared with the real tag corresponding to the picture, the accuracy rate is 72.2%. The open platform of emotion recognition based on deep learning enables external programs to use emotion recognition services through open API, providing more convenient and faster development methods for developers, thus promoting the wide application and promotion of emotion recognition technology. The next step is to further strengthen the robustness of the platform, and constantly optimize the emotion recognition algorithm to improve the recognition accuracy.

Target detection in 3D vision holds an important role in the visual understanding system of self-driving cars, and the use of multiple sensors can effectively overcome the limitations of a single sensor and enhance the accuracy and reliability of the intelligent vehicle's perception of the surrounding environment. So, the research generally adopts the method of fusion of multisensor information to detect the target. In this paper, the distance measurement of the target is accomplished through data fusion between the LiDAR point cloud data and images. Furthermore, the proposed distance measurement technology based on R-tree is based on target recognition with images captured by the on-board camera. A leakage detection removal technique based on radar point cloud enhancement is proposed to solve the problem of serious interference with the recognition of deep neural networks caused by the poor imaging quality of photos taken by vehicle-mounted cameras in special scenes and conditions.

Due to the low accuracy and poor recognition performance of traditional methods in end-to-end OCR text recognition of power equipment nameplates, an end-to-end OCR text recognition method for power equipment nameplates based on image enhancement and template matching is proposed. Calculate the similarity of end-to-end OCR text data collection on power equipment nameplates and extract end-to-end OCR text features on power equipment nameplates; Based on actual node deployment requirements and standards, combined with image enhancement and template matching technology, OCR text feature recognition nodes are deployed to achieve end-to-end OCR text recognition on power equipment nameplates. The experimental results show that the proposed method has a maximum recognition accuracy of 99.67%. The difference between the recognition results and the actual situation is small.

In order to enrich the electric power training based on virtual reality technology, this paper focuses on the design and implementation of gesture recognition function. Through the configuration of HTC Vive locator and Gabor gesture recognition algorithm, plus data glove and VR headset for auxiliary use, to complete the natural and intuitive human-computer interaction in VR. Among them, the HTC Vive locator can locate the user's position in real time in the virtual environment, and the Gabor gesture recognition algorithm is to analyze the user's hand movements and posture. The simulation test with the students of Hainan Electric Power Training School verifies the accuracy of the design of the scheme, which can improve the training effect and learning results.

According to the physical structure of Trench Gate IGBT and the distribution characteristics of excess carrier in the base region, an improved on-state model considering the two-dimensional distribution of carrier is proposed on the basis of improving the shortcomings of the existing model. In this model, the whole IGBT base region is divided into several regions, and the continuity equation is solved by different boundary conditions, so that the distribution of excess carriers in the base region can be accurately reflected and more accurate simulation results of V-I characteristics can be obtained. Finally, the accuracy of this model is verified by experiments.

Chinese medical named entity recognition is a crucial task in information extraction within the medical field. Its primary objective is to extract entities of medical significance, such as diseases, clinical manifestations, and drugs, from medical texts. Due to the specialized nature of the medical domain, the grammatical structure of medical texts is often more complex than that of other texts, frequently containing a plethora of abbreviations and terms. To enhance the effectiveness of Chinese medical entity recognition, this paper introduced a pre-trained BERT-BiLSTM-CRF model, the CHIP2020 medical text dataset was chosen for the experiment, including 9 types of entities, such as diseases, clinical manifestations, drugs, medical equipment, medical procedures, body parts, medical examination items, microorganisms, and departments. BERT effectively utilizes context information to comprehend the semantics and relevance of words in sentences, effectively resolving the challenge of word polysemy and exhibiting strong language representation capabilities. Additionally, BiLSTM is employed to extract features, efficiently capturing the sequence information and context between words. Finally, CRF is utilized to convert the probability sequence output from BiLSTM into the optimal annotation sequence. Experimental results demonstrate that, compared with three other models, the proposed method yields improved accuracy in named entity recognition on the CHIP2020 dataset, with an impressive F1 value of 64.93%. This approach effectively extracts medical entities.

Blind image restoration is a challenging and valuable subject in the field of computer vision. In this paper, under the framework of Total Variation (TV) regularization term, the bright and dark channel regularization term is introduced into the multi-frame image blind restoration algorithm, aiming at overcoming the shortcomings of low contrast and obvious ringing effect of the restored image on the basis of retaining the advantages of TV regularization. The bright channel regularization term can estimate the blur degree of the degraded image more accurately, while the dark channel regularization term can better resist noise interference. The split Bregman iterative algorithm is used to solve the objective function, and the details of the restored image are compensated in the iterative process, and further. The experimental results show that the algorithm in this paper has a certain deblurring effect, and the peak signal-to-noise ratio of the restored image is improved by about 0.53 dB to approximately1.5 dB, and the structural similarity is improved by about 0.12 to approximately 0.47, with better visual effects.

Aiming at drivers' dangerous driving behavior monitoring and health monitoring of drivers, this paper designs an intelligent steering wheel which can monitor dangerous driving behavior and a steering wheel sleeve which can monitor physical health. The MTCNN model is primarily used to obtain the driver’s face image in real time. The PFLD algorithm was used to obtain the facial model positioning feature points, and the degree of driver's fatigue was determined by combining the relevant parameters. The fatigue algorithm proposed in this paper can improve the effectiveness and accuracy of the monitoring. Then, according to the LSTM network model, 11 groups of key point information of the human body are obtained, and the human motion track is identified, and then combined with the facial information, so as to complete the judgment of driving behavior such as drinking water, smoking and walking. Through the PPG and ECG fusion algorithm based on LSTM, the reliability of the system to collect vital signs such as body temperature, blood pressure, heart rate and blood oxygen of the driver is improved. It was experimented that the system could monitor a driver's driving behavior in real time and consider its health management.

At present, there are many difficulties in the judgment of depression. Under the guidance of experts, this study selected a total of 26 depression features from four aspects: self-evaluation, psychological state, social support and physiological signs. Based on C4.5 decision tree algorithm, a depression prediction model was constructed. The experiment showed that the prediction accuracy was above 69%, which could play a role in the preliminary diagnosis of depression.

Although there are many differences in writing and vocabulary between the late Qing Dynasty to the early Republic of China (QDRC) of Chinese and modern Chinese, there is not much difference in grammar, semantics, etc. It is a transitional stage between ancient Chinese and modern Chinese. We seem more willing to refer to this stage of Chinese as the pre modern Chinese period. This paper refers to the method of discovering new words and phrases in the natural language processing (NLP) field, aiming to explore a new method of discovering domain specific words (DSW) in Chinese texts from the QDRC period. Through the comprehensive application of corpus linguistics and computational linguistics, we have successfully constructed a multi-level analytical framework to reveal the unique lexical characteristics of Chinese texts in this period. First, this paper establishes a small Chinese corpus at the QDRC period, and uses bigram and trigram of N-gram to preprocess the texts, including word segmentation, parts of speech tagging, etc. The word frequency, pointwise mutual information, left and right entropy of bigram and trigram fragments were calculated, and 909 key words with special status in social, political, cultural and other fields in the QDRC were finally identified through manual proofreading. This article validates the effectiveness of the proposed method and discovers some distinctive Chinese words in the QDRC that have not been previously studied in the experiments.

Recommendation systems as one of the main products of today's Internet companies, as well as in view of the large number of current recommendation algorithms used in various mobile platforms. This paper focuses on the horizontal and vertical exploration of the model of the popular recommendation algorithm AFM. A cross-sectional comparison is made with two other recommendation models, FM and NFM, to demonstrate that AFM is superior in terms of some model evaluation metrics. In addition, the AFM recommendation model is improved by setting different temperature parameters and different feature interaction degrees to demonstrate the performance of AFM in terms of recommendation accuracy in different dimensions.

Radar target recognition technology is the intersection of radar technology and pattern recognition. It is a technique to extract the information and features related to the target characteristic from the High-Resolution Range Profile (HRRP) and to judge the target attribute. Due to the complexity of radar target recognition algorithm and large amount of data, most of the current research is still focused on algorithm, and few can be implemented on hardware. With the development of microelectronics, System on Chip (SoC) has been widely used. Its powerful data processing capability makes it possible to realize the hardware implementation of radar target recognition. This paper mainly introduces the methods of feature extraction and classifier, and then implements a complete radar target recognition system based on the hardware platform of RF-SOC chip. Experiments show that the classification accuracy of ground targets such as tanks, radar stations and armored vehicles is over 80%.

Emotions are important signals in human communication in daily life and are also considered a crucial factor in human-machine interaction. Conducting emotion analysis based on text data holds significant importance for expanding the scope of machine applications and enhancing interactivity. In this paper, we conduct emotion analysis research using the public Twitter dataset, focusing on two widely used models in the current NLP field for emotion analysis: the Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM). Additionally, we explore the application of large pre-trained language models, represented by ChatGPT, in the domain of emotion analysis. The experimental results indicate that while CNN and BiLSTM models achieve relatively high accuracy in supervised emotion analysis, ChatGPT demonstrates superior performance in unsupervised learning for emotion analysis. Furthermore, the exploration of emotion analysis based on ChatGPT reveals some previously unnoticed issues.

Electronic nose (E-nose) technology, inspired by the biological olfactory system, has exhibited substantial promise across diverse applications due to its effective odor detection and identification capabilities. This paper presents a novel approach that leverages a bioinspired olfactory bulb model for processing data within E-nose systems. In this study, the complex neural circuits within the olfactory bulb were simulated using a bioinspired model. This method directly harnesses the inherent feature extraction capabilities of the olfactory bulb model, eliminating redundant steps and thereby enhancing classification accuracy. To evaluate the performance of this approach, ten types of toxic gases were classified using Support Vector Machine (SVM), Extreme Learning Machine (ELM), and Linear Discriminant Analysis (LDA) classifiers. The proposed bioinspired olfactory bulb model was compared to traditional data preprocessing methods. Experimental results clearly demonstrate that the bioinspired olfactory bulb model significantly improves classification accuracy compared to traditional methods.

Named Entity Recognition (NER) for chemical experiment operations can not only extract key information and automatically generate operation instructions in the field of automated synthesis but also facilitate chemical experiment personnel in analyzing literature data more efficiently. In this paper, we propose a NER model that combines multiple layers of BiLSTM and IDCNN in parallel, based on the Bert pre-trained model. By adjusting the number of BiLSTM and IDCNN modules at each layer, we can extract more contextual information and local feature for different datasets, and subsequently generate entity labels using a Conditional Random Field (CRF) layer. The experimental results indicate that the model achieves an F1 score of 0.9174 in the constructed dataset, surpassing existing algorithms.

By improving the Mask R-CNN model, the performance of the model is improved for recognizing and segmenting apples in the orchard that are in natural environments such as different light, different weather and overlapping occlusion. Firstly, the dual-channel SE module is introduced into the feature extraction network, secondly, the anchor frames are modified according to the actual exposed proportion and area of the apples, and finally, the NMS is improved to Soft-NMS to enhance the model's recognition ability for apple identification and segmentation in natural environments. The results of the experiments using apple images from real orchard environments: Pixel Acc was 98.4%, mean intersection and Mean IoU was 78.8%, class intersection and Class IoU was 59.4%, and Class Mean Acc was 84.8%. The results show that the improved model in this paper gives better results than the original Mask R-CNN model for apple recognition and segmentation, as well as some other models that are widely used.

The selection of mining method is a comprehensive judgment with multi-objective and multi-hierarchy. The traditional selection methods are not scientific enough for traditional decision-making. Here, based on the application of the combined analytic hierarchy process and technique for order preference by similarity to ideal solution (AHP-TOPSIS) on various possible mining method. A model of fuzzy hierarchy analysis was built up to evaluate and determine the best mining method. Our results proved that the AHP-TOPSIS is a practical, reliable and scientific method for the determination of mining method.

In real life, image forgery techniques such as stitching, copying, moving, deleting, and enhancing operations are rampant, causing serious social harm. We propose an image forgery detection called LaTe-Net. LaTe-Net does not require additional preprocessing or postprocessing and can detect images of any size and image forgery operation type at the same time. In this paper, we designed a texture enhancement module that focuses on local texture features and amplifies subtle pseudo shadows in shallow features to detect forged images by identifying local abnormal features. In addition, we perform forgery localization through local abnormality detection to capture local abnormalities. Finally, we conducted comparative and visual experiments on different public datasets, and the experimental results demonstrate that LaTe-Net has good performance and generalization properties for different types of forgery operations.

In order to solve the problem that existing session recommendation methods ignore the temporal relationship between items and do not consider the correlation between the items of interest and all items, which leads to poor recommendation results. A novel session recommendation model based on Graph Neural Network (GNN), Long- and Short-Term Memory network (LSTM) and attention mechanism is proposed. The GNN and LSTM are used to capture the complex dependencies and temporal relationships between items to obtain item embeddings; the global embeddings are generated by combining multi-headed attention and soft attention to accurately represent global preferences; the interest attention mechanism is introduced to generate interest embeddings to activate interest item relevance; finally, the current embeddings are combined to obtain the final session representation and predict the next click. The model is compared with the existing benchmark model on three public datasets and the results are all improved, which proves that the proposed method is very effective.

Badminton is a sport for young and old alike. To capture the dynamic path of badminton quickly and accurately, a badminton robot needs to be designed for real-time dynamic tracking of badminton moving at high speed. In this paper, we analyze the effects of aerodynamic forces and gravity on badminton flight in the air. The main factors affecting the flight trajectory of badminton are analyzed using experiments. The badminton flight trajectory captured by a high-speed camera is used to establish a steady-state flight model as well as a spin model for badminton. In the process of badminton trajectory prediction, because there are model errors as well as measurement errors, particle filtering is an inference model based on Monte Carlo sampling, which does not require a priori information of state transfer and is more conducive to badminton state estimation. Therefore, in this paper, particle filtering is chosen to track the path of badminton. After experimental verification, the algorithm can track the motion path of badminton very well.

This paper investigates the issue of cold chain safety during major outbreaks. By constructing an evolutionary game model, the dynamic game process between government, cold chain platforms and logistics firms is analyzed. Combined with system dynamics, simulation analysis of the dynamic game between the three parties’ strategic choices is conducted. The study shows that the behavioral strategies of all three parties have a certain degree of interactive influence. Regardless of the initial state, the system eventually evolves into a stable state of (1,1,1). Furthermore, the three parties involved in cold chain prevention and control are highly sensitive to external variables, and the final strategy is influenced by multiple factors. To effectively prevent and control the epidemic, the government should play a leading role and take various measures to stimulate the initiative of the participating parties.

Based on the principle and system model of orthogonal frequency division multiplexing with index modulation (OFDM-IM) technology, the performance of Maximum Likelihood (ML) detection algorithm and Logarithmic Likelihood Ratio (LLR) detection algorithm are compared by matlab simulation modeling analysis, and the computational complexity is presented. At the same time, the effects of different number of subcarriers, different activator carriers, different modulation modes and different modulation orders on the performance of system bit error rate are analyzed. Following the principle of mutual conversion of validity and reliability, it provides a reference basis for the compromise between the spectrum efficiency and the system performance.

An essential function of video surveillance systems that are widely used in fields like public safety is automatic anomaly detection. To efficiently employ spatial-temporal data, the system in this thesis combines geographical branching and temporal branching in a single network. The network has a residual self-encoder structure that, along with a DMA-based encoder and a multi-stage channel attention-based decoder, first uses a spatial-temporal wavelet analysis module for multi-frequency decomposition of video content to improve the model's understanding of high-frequency detail information and processive motion. The model learns more effectively thanks to the implementation of a channel attention mechanism that takes advantage of the channel correlations between features. Finally, training is carried out without supervision. While context dependency is retrieved from the channel attention module, the time-shifted technique is employed to take advantage of temporal aspects. The UCSD dataset is utilized to assess the system's performance. The outcomes demonstrate that our network outperforms current approaches.

Optimizing the energy use of adjustable load and improving the reliability and economy of building power supply are the key issues of building power supply system scheduling. Aiming at the problem of load energy optimization of public buildings, this paper proposes a building load energy optimization method based on the improved Grey Wolf Optimizer. Firstly, an optimal scheduling model was established based on the energy supply system of public buildings. Secondly, the teaching factor is introduced into Grey Wolf Optimizer (GWO) to improve its iterative updating mode, and to improve the searching ability of the algorithm. Finally, the improved Grey Wolf Optimizer is used to solve the economic optimization model of public building energy supply system, and the optimal results of day-ahead operation and cost of the public building energy supply system are obtained. Simulation examples show that compared with the particle swarm optimization algorithm and Grey Wolf Optimizer, the improved Grey Wolf Optimizer (IGWO) proposed in this paper has obvious advantages in terms of fast convergence and global optimization, and the proposed method can more effectively reduce the operating cost of the energy supply system of public buildings.

As we all know, the objects in the images taken by Unmanned Aerial Vehicle (UAV) are relatively small, while our naked eyes are able to extract the information almost instantly, even from far away, image resolution and computational resources limitations make detecting smaller objects a genuinely challenging task for machines. We propose an algorithm based on YOLOv5s with small computational resources and high accuracy, so as to be applied to edge detection devices such as unmanned aerial vehicles. By simplifying the depth of the feature extraction network and adjusting the size of the feature map of the detection head, the target in the image taken by UAV can be accurately identified. In the end, we reduced the number of parameters by 70% at the expense of a little accuracy, while improving accuracy by 15.25%, or 5.2 percentage points, over the baseline.

In aviation accident classification and prediction, there are numerous factors that contribute to accidents, and these factors have intricate relationships, making it difficult to classify accident types accurately. Current classification methods for aviation accidents face issues like inadequate data processing and failure to eliminate invalid samples, resulting in complex data and low accuracy. To address these problems, a classification method called Lasso SVR is proposed for aviation accident inducing factors. It aims to accurately classify aviation accident data. The method involves constructing an influential factor matrix containing eight representative factors and five accident grades. The Pearson correlation coefficient is then calculated to determine the data correlation, and dimensionality reduction is achieved using the Lasso minimum angle regression method. By optimizing the penalty coefficient, the weights of accident influencing factors are obtained. Support vector regression technology is employed to process the influential factor matrix and achieve accident classification, along with in-depth characteristic analysis. Experimental results show that the Lasso SVR model reduces computational complexity and data complexity, enabling faster approximation of linear relationships in the data. It performs well in processing aviation accident data, improving accuracy and providing better data fit compared to traditional SVR models. Additionally, the Lasso SVR model demonstrates superior performance, particularly in terms of fitting linear kernel functions, when compared to other kernel functions.

Our proposed approach for predicting the production capacity of gas wells in carbonate reservoirs involves the utilization of a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network integrated with a fusion attention mechanism, addressing the significant errors in traditional empirical formula and mathematical analytical methods for carbonate reservoir gas well production prediction. By utilizing deep learning methods, we can accurately capture the complex relationships and nonlinear patterns among a large volume of data and extract valuable information from extensive datasets, enabling precise predictions of carbonate reservoir gas well production capacity. The incorporation of the attention mechanism into the Bi-LSTM network architecture enhances not only the ability to handle input subsets but also the capability to process time series data. We apply this model to predict the daily gas production of carbonate reservoir gas wells, and the results demonstrate superior prediction performance compared to individual Bi-LSTM and LSTM models.

Based on issues such as damage leading to the failure of the three-segment code and the inability to automatically sort, we propose a rapid recognition solution for express delivery waybills based on YOLOv5_obb. Compared to YOLOv5, this method introduces a rotated rectangle regressor that handles image angle correction and image segmentation together. Experimental results demonstrate that this approach achieves an accuracy rate of up to 99.5% in the extraction of express delivery waybills, effectively enabling high-speed extraction.

Network is a double-edged sword, which brings great convenience and some risks at the same time. Then intrusion detection, as a proactive and defensive technical means, plays a very important role in protecting information systems. Traditional intrusion detection systems have the problems of low detection accuracy and poor real-time performance, while the existing network intrusion detection data is not balanced, so this paper proposes an adaptive synthesis algorithm (ADASYN) oversampling and class balancing data balancing algorithm and an intrusion detection method of LightGBM combined with Bayesian optimization algorithm, aiming to solve the above problems. First, data preprocessing we encode the data with labels as well as unique heat encoding to better pass into the model for processing, after that we increase a few classes of data by ADASYN oversampling to make the number of samples in each class relatively balanced, and finally the features are passed into LightGBM for learning and training, and finally validated in the NSL-KDD dataset for multi-class classification experiments, and the results show that compared with other traditional algorithms, all data are improved and show better performance.

Wearing masks can prevent infection of COVID-19 to a large extent. It is tedious and inefficient for COVID-19 high-risk groups to rely on manual supervision and reminding to wear masks. However, some mask detection systems have the problem of high detection accuracy but low detection rate or high detection rate but low detection accuracy now. As a result, this paper proposes a mask wearing detection model based on YOLOv7 algorithm, which randomly extracts 3600 images from the public RMFRD dataset and uses Mosaic data augmentation technology for experiment. The experimental results show that the detection in the test set achieved an accuracy of 96.43%, a detection speed of 140f/s, and a recall rate of 94.43%. The model achieved good results in both speed and accuracy when performing mask wearing detection tasks.

In this paper, a monocular vision measurement method based on feature detection and extraction and PnP algorithm is used to solve the relative pose of the spatial target, which is used to quickly and accurately determine the positional relationship between the two targets in space. Based on the structure of the spatial target, the pixel coordinates of four coplanar feature points and two non-coplanar feature points on the spatial target are extracted to the image using image processing methods, and then combined with the PnP problem based on six feature points is solved by combining the dimensional information of the spatial target, and the relative positions of the camera and the spatial target are solved. The experimental analysis shows that the spatial target position calculated by this method is close to the real result, which verifies the effectiveness of this method.

In order to improve the encoding performance of 3D videos, the 3D-HEVC encoding standard has also added new stereo video encoding technologies such as interview prediction and depth map encoding. However, excellent encoding performance is achieved through significant computational complexity. In order to reduce the computational complexity of the 3D-HEVC encoding process, this paper proposes a 3D-HEVC fast encoding algorithm based on Bayesian decision-making. This algorithm uses machine learning methods to extract features that meet specific relationships and uses these features to learn and establish classification and prediction models. The experimental results show that compared with traditional 3D-HEVC encoding algorithms, the machine learning method based on Bayesian decision-making can effectively improve the video coding performance of 3D-HEVC, with a reduction of 23.28% in total encoding time and 39.09% in deep mapping encoding time, respectively.

In this paper, a model for STIRPAT-based fusion predictive analytics is proposed by building ARIMA time series model, BCC model, data envelopment analysis model, system dynamics model and STIRPAT model. The model solves the problem of predicting and analyzing heterogeneous data. This paper uses data such as IHSMarkit analysis reports and related patents to confirm the validity of the model. Finally, the model successfully predicts the time to achieve carbon neutrality in the Yangtze River Delta region and the amount of automobile ownership.

SysML (System Modeling Language), a standard modeling language for system engineering, is widely used in the field of aerospace engineering. Model transformation technology has proven to be effective in accelerating the design process, improving the modeling efficiency, and reducing the workload. In this paper, a new method for the model transformation is proposed considering the element mapping from activity diagram to sequence diagram. Firstly, the relationships between the messages of sequence diagram are established, based on which the mapping rules of the sequence diagram and the activity diagram are introduced considering various relationship categories. Secondly, according to the states of nodes and connectors presented in the activity diagram, the transformation model from the activity diagram to the sequence diagram is achieved by determining the messages and the interaction objects of the sequence diagram. Finally, to validate the feasibility of the proposed method, a comprehensive case study is conducted through an aircraft flap control system, utilizing the self-developed software.

A fusion detection method using multi line lidar point cloud detection combined with image recognition analysis is proposed to address the hazards of obstacles such as railway landslides and falling rocks. After clustering based on Euclidean distance, it is proposed to apply real scene restriction information to filter out interference clustering, while combining inter frame information to greatly reduce the false alarm rate of point cloud detection. Image detection uses YOLO-V5 for object detection and proposes to use a whole target sequence to achieve target information matching. On the basis of unifying the coordinate system of radar and camera, a "Intersection small ratio" method is proposed to achieve the fusion of camera and radar detection results. The research results indicate that the obstacle detection method combining point cloud and image can accurately and reliably detect people, trains, and various types of obstacles within the region, providing new technical support for railway line state perception.

Object detection is a fundamental task in computer vision that has garnered significant attention and made remarkable strides in recent times. Most object detection tasks are designed only for visible images, but the fusion between visible and infrared images can play a complementary role in information so that object detection can play a better effect in low light, bad weather, and drone shooting environment. Although some progress has been made in visual and infrared image fusion object detection, and a series of algorithms have been proposed, there is no article on the comprehensive introduction of RGB-Infrared image fusion object detection. Therefore, we summarize the current status of object detection based on RGB-Infrared fusion and discuss the limitations of current techniques and potential research directions in the future. We hope that this article can help researchers quickly understand the field, then provide certain suggestions and references for relevant scholars.

In recent years, the increasing number of motor vehicles in China and the limited road infrastructure resources have resulted in significant challenges. Intersections play a vital role in regulating urban traffic, and effective dynamic signal timing is crucial for improving intersection efficiency and overall traffic network performance. This paper focuses on studying the optimization of dynamic signal timing for a single intersection. It proposes a dynamic signal timing model based on multi-objective optimization, establishing an optimization function that considers capacity, delay time, and the number of stops as key performance indicators. To solve the model, an improved hybrid algorithm combining Genetic Algorithm and Particle Swarm Algorithm is employed. The model's performance is evaluated through joint simulation experiments using PTV-VISSIM and MATLAB software. The experimental results demonstrate that the proposed timing optimization model outperforms the classical Webster model. Specifically, it reduces the maximum queuing length of vehicles at the intersection by 49% and decreases fuel consumption by 25%. Consequently, the proposed model significantly improves intersection control effectiveness.

SysML (System Modeling Language) has been widely used in the aviation field for the architecture design and the system modeling. However, the modeling of complex systems based on SysML requires significant time and effort. To reduce the workload of SysML modeling and improve the efficiency when designing, a new method is proposed to transform the SysML activity diagrams into the state machine diagrams. First, the activity diagrams and the state machine diagrams are defined. Second, mapping rules to transform the activity diagram elements into the state machine diagram ones are established, based on which the predefined state machine diagram is generated. Third, the diagram is transformed by modifying or adding elements as needed. Finally, the proposed method is demonstrated through a numerical case comes from the aircraft flap control system.

With the rapid growth of the digital economy in the Internet era, traditional crimes have expanded into the realm of cyberspace, leading to the emergence of telecom network fraud crimes. This paper specifically focuses on telecom network fraud crimes. By utilizing NVivo software, a qualitative analysis was conducted on 53 adjudication documents in database to develop a model aimed at comprehending and combating telecom network fraud crimes. Based on the analysis of the model, the following conclusions have been drawn there is a need to strengthen the collection and utilization of evidence, enhance anti-fraud education, and improve the gathering of basic information on telecom network fraud.

To solve the current target detection algorithms in remote sensing images, which have problems of small target detection accuracy is not high and easy to produce error detection and omission detection under the target dense and complex background, this paper introduces two enhancements rooted in the YOLOv7 algorithm. First, the ACmix attention mechanism is added after the backbone network of YOLOv7 algorithm, and second, drawing on the ideas of Soft-NMS algorithm and DIOU-NMS algorithm, CHSD-NMS algorithm is proposed, which improves the base NMS algorithm by softening the strategy of eliminating the candidate box and replacing the IOU with the DIOU to change the traditional NMS algorithm violent removal of candidate boxes. To evaluate the efficacy of the proposed algorithm in this study, experiments are conducted on the remote sensing image dataset DOTA and compared with the basic YOLOv7 algorithm. The experimental results indicate that the algorithm's average accuracy improves by 2.9% with the incorporation of the ACmix attention mechanism at IOU=0.5, a further enhancement of 3.2% is achieved by substituting the traditional NMS algorithm with the CHSD-NMS algorithm, and the overall average accuracy of the algorithm in this study increases by 5.2% with the implementation of both improvements compared to the conventional algorithm.

Wordle is a puzzle game where players guess a hidden five-letter word in six attempts or less. This article conducts research on word attributes based on the game's requirements to enhance gameplay. Through correlation analysis between word attributes and percentage scores, we propose an ARIMA model to predict variations in player numbers, an ARIMA-GBDT predictive model to forecast attempt results, and a word difficulty division model based on FCM clustering. Experimental results demonstrate that the ARIMA (1,1,0) model, predicting player number variations, achieves better prediction results (DW 1.89617 on a 15% validation set). Furthermore, the ARIMA-GBDT predictive model outperforms both the ARIMA and GBDT models, and the FCM clustering model achieves a ST value of 0.8360 with an accuracy of 84.21% (based on 1000 samples), showcasing the best performance.

This research reviewed the rapidly developing field of abnormal behavior detection in video surveillance. The emergence of deep learning techniques, especially in video feature extraction, makes up for the shortcomings of traditional methods. In terms of abnormal behavior detection, unsupervised, supervised, and weakly supervised methods have different advantages and disadvantages. At present, the weakly supervised method is popular in this field, and the highest AUC under the UCF dataset reaches 86.98%. Benchmark datasets play a crucial role in evaluating the performance of algorithms. Future research will focus on addressing challenges related to complex scene dynamics, occlusions, and real-time processing. Integrating multiple sensing modalities, transfer learning, deep learning techniques for feature extraction, and leveraging spatiotemporal graph-based methods are important directions for improving surveillance systems in anomaly detection.

As an active microwave imaging system, Synthetic Aperture Radar (SAR) has the characteristics of long-distance detection, high-resolution, all-time and all-weather imaging, so it is widely deployed. As an active jamming means, radio frequency interference can reduce the quality of SAR images by transmitting radio frequency signals to form a blanket effect on SAR images. In order to recover the original image from the interfered image, this paper firstly constructs a signal model for common RFI, secondly constructs a RFI dataset based on the above signal model, and lastly proposes an interference suppression method based on UNet, adopting encoder network for data dimensionality reduction, using decoder network for image reconstruction, and adopting skip connection between encoder and decoder to increase information exchange. The final experimental results show that compared to the Dncnn, the proposed method has an absolute improvement of 2.1dB in PNSR.

In recent years, as short video content has been recognized by the public and the popularity of short video emerging ways to improve, short video playback technology is faced with a variety of device models, complex audio and video streaming transmission, sensitive network environment, and other status quo. The traditional short video function based on native development has been unable to effectively deal with some specific video playback scenarios, especially not for "enterprise-level" mobile comprehensive short video needs. In this paper, the solution is based on HTML5 technology to achieve lightweight short video playback, through the design of multi-player mechanism, to solve the traditional HTML5 single-player lack of preloading capabilities, so that the application has the same user experience as the native video player, but also through the HTML5 way to achieve the flexibility of the application release. Finally, this solution realizes a lightweight short video playback solution based on the multi-player preloading mechanism, improves the smoothness of short video playback and conducts a comparative analysis with the existing solutions, which verifies the simplicity and flexibility of the solution proposed in this paper.