Content of Engineering and Applications in our journal

Published in last 1 year |  In last 2 years |  In last 3 years |  All Please wait a minute... For Selected: Download Citations EndNote Ris BibTeX Toggle Thumbnails Select Object Detection Within 3D Point Cloud via 2D Convolution Neural Network LI Xiaoli, WANG Le, DU Zhenlong, CHEN Dong Computer Engineering and Applications    2025, 61 (23): 297-304.   DOI: 10.3778/j.issn.1002-8331.2409-0082 Abstract （19）      PDF（pc） （3009KB）（20）       Knowledge map    Save Lidar has been initially applied in autonomous driving and industrial automation, generating vast amounts of point cloud data for scenes and objects. These point cloud data are characterized by high dimensionality and irregularity, and require computationally expensive 3D convolution in existing deep learning models, leading to high spatio-temporal complexity and hindering online application. Addressing the limitations of traditional network models in processing point cloud data, this paper proposes a 3D point cloud object recognition method based on 2D convolutional neural networks. The proposed method statistically regularizes irregular point cloud data into pillars, utilizes convolutions and pooling to extract features from clusters of pillars, converts the 3D point cloud data into 2D image-like features, and employs 2D convolutional neural networks to extract multi-scale latent features from multiple receptive fields. The decoder network then identifies objects within point cloud based on locations, orientations, and object types. Experiments are conducted on Ascend Atlas 200DK edge devices, achieving a single inference time of 291?ms. Compared with traditional point cloud object detection networks, the proposed method outperforms VoxelNet, F-PointNet, and Second by 14.7, 13.2, and 3.4 times, respectively, in terms of performance gains. On the KITTI dataset, the average precision exceeds that of the second-best algorithm by more than 2.3%  compared with 14 other point cloud object detection algorithms, including ContFuse. Ablation studies focusing on 2D convolutions and attention mechanisms reveal improvements of 50.9% and 5.37%, respectively, in model size and inference accuracy. The experimental results demonstrate that the proposed method can efficiently, robustly, and accurately detect objects within point cloud data. Reference | Related Articles | Metrics Select Method for Foreign Object Detection on Transmission Lines Using Monocular Depth Estimation HU Guangyi, HAN Jun, NI Yuansong, WANG Wenshuai, CHEN Keyu Computer Engineering and Applications    2025, 61 (23): 305-315.   DOI: 10.3778/j.issn.1002-8331.2407-0097 Abstract （14）      PDF（pc） （3188KB）（22）       Knowledge map    Save To address background false positives and object misses in detecting foreign objects on transmission lines, a method using monocular depth estimation is proposed. The multi?level feature fusion depth estimation network (MFFDepth) integrates semantic information from multiple feature levels within the encoder and introduces a coordinate attention module in the skip connections between the encoder and decoder, enhancing global depth perception in complex scenes. Based on the predicted depth map, depth value clustering is used to obtain the foreground image and foreground depth threshold. The YOLOX object detection network, combined with the foreground depth threshold, excludes background false positives. The DeepLabv3+ semantic segmentation network, combined with the depth foreground image, addresses the issue of foreign object detection omission. Finally, the results from these two combined detection modules are fused to improve overall detection performance. Experimental results show that the proposed method achieves an accuracy of 92.9% and a recall rate of 95.8%, which are improvements of 1.4% and 8.3%, respectively, compared to the original YOLOX algorithm, effectively enhancing the detection of foreign objects on transmission lines. Reference | Related Articles | Metrics Select Reinforcement Genetic Algorithm for Path Planning of UAV Power Inspection with Nest Charging LIANG Chenlei, LUO He, JIANG Ruhao, YIN Youlong, LIN Shizhong, WANG Guoqiang Computer Engineering and Applications    2025, 61 (23): 316-328.   DOI: 10.3778/j.issn.1002-8331.2407-0323 Abstract （21）      PDF（pc） （1362KB）（33）       Knowledge map    Save Aiming at the path planning problem of UAV power inspection with the machine nest as the charging station, a mathematical model is constructed to minimize the total time of UAV task execution, and a reinforcement genetic algorithm is designed to solve the problem. In this algorithm, a population initialization operator based on greed and a feasible solution generation operator based on split are proposed, and the parameter tuning process of genetic algorithm is modeled as a Markov decision process, and a dynamic tuning strategy of cross probability and mutation probability is designed based on double Q-learning. In numerical experiments, the results of comparison with Gurobi solver, classical genetic algorithm, genetic algorithm based on elite retention and differential evolution algorithm show that the algorithm has significant advantages in solving quality and solving speed. At the same time, in the case analysis, the comparison with the existing inspection strategy further verifies the application effect of the algorithm in the actual scene. Reference | Related Articles | Metrics Select Real-Time Dynamic Visual-Inertial SLAM Algorithm Integrated with Gaussian Mixture Filtering WANG Yudong, WU Helei, XU Xuesong Computer Engineering and Applications    2025, 61 (23): 329-339.   DOI: 10.3778/j.issn.1002-8331.2408-0379 Abstract （23）      PDF（pc） （12747KB）（45）       Knowledge map    Save Regarding simultaneous localization and mapping (SLAM) with poor robustness, low accuracy and weak real-time performance in dynamic environment, a visual inertial SLAM algorithm integrating Gaussian mixture filtering is proposed. Initially, the visual feature point displacement is calculated by designed spatial sifter from the visual points coordinate and the camera prior rotation estimated by error state Kalman filter (ESKF) used on inertial measurement unit (IMU). Subsequently, the Gaussian distribution of feature points is sieved to obtain the initial expectation and its variance, and the Gaussian mixture model is introduced to optimize each Gaussian distribution and generate the corresponding feature point clusters. Afterwards, the optimal static cluster filtering strategy is proposed to obtain the stable static feature point cluster and estimate the accurate camera pose. The experimental results based on TUM-RGBD and VCU-RUI dynamic landmark dataset show that the proposed method has better performance than VINS-Mono and its improvement in most dataset. It achieves an average enhancement of 92% in root mean square error of absolute trajectory error compared to VINS-Mono, meets real-time requirements, and has reference and potential applications for SLAM research and autonomous robot navigation. Reference | Related Articles | Metrics Select Research on Joint Distribution Optimization of Multi-Center and Multi-Vehicle Electric Trucks Under Time-Varying Road Network GUO Jiawei, HUANG Zhipeng, JIA Jinxiu, MA Xiaotian, LI Jianguo, YE Binbin Computer Engineering and Applications    2025, 61 (23): 340-350.   DOI: 10.3778/j.issn.1002-8331.2408-0441 Abstract （15）      PDF（pc） （1603KB）（11）       Knowledge map    Save In the transformation of the logistics industry from high carbon emissions to green and low carbon, electric trucks are favored in the field of logistics distribution. However, considering the uneven temporal and spatial distribution of traffic impedance in urban road networks and the nonlinear characteristics of battery charging, traditional static vehicle routing optimization is difficult to meet actual needs. In order to improve the delivery efficiency of electric trucks in time-varying road networks, a mixed integer programming model with the goal of minimizing the comprehensive delivery cost is constructed by comprehensively considering factors such as multi-center and multi-model joint distribution strategy, partial charging strategy based on nonlinear charging function, time window, load and service time window. An improved K-means clustering method and a simulated annealing algorithm with memory function are designed to solve the model. Taking some logistics parks in Shanghai as an example to verify the effectiveness of the model and algorithm, the results show that the distribution cost difference between peak and non-peak hours is about 5.7%. Compared with the single vehicle distribution scheme, the cost of the multi-vehicle joint distribution scheme is reduced by about 5.4%. The cost of the partial charging strategy is about 5.4% lower than that of the full charging strategy. The research results provide a reference for logistics enterprises to further optimize the distribution scheme of electric trucks under urban time-varying road network. Reference | Related Articles | Metrics Select School of Computer and Electronic Information, Guangxi University, Nanning 530004, China LI Zhijun, CHEN Qiulian Computer Engineering and Applications    2025, 61 (23): 351-359.   DOI: 10.3778/j.issn.1002-8331.2408-0453 Abstract （17）      PDF（pc） （1403KB）（17）       Knowledge map    Save In view of the disadvantages of heuristic intelligent algorithm bat algorithm, which is easy to fall into local optimization and insufficient optimization ability, an improved discrete bat algorithm is proposed with the goal of minimizing the maximum completion time. Firstly, the paper combines the selection of local minimum time machines and random selection machines to initialize the population, improving the quality and diversity of the initial population. Secondly, from the perspective of process arrangement and machine selection, selection, superposition, crossover operators, and forward and reverse learning operations are designed to improve the position update mechanism, and use six neighborhood structure operations based on operation arrangement and machine selection to optimize the variable neighborhood search strategy and enhance the algorithm’s ability for global and local search. Finally, the experimental simulation results of benchmark examples show that the improved discrete bat algorithm has better optimal performance. Reference | Related Articles | Metrics Select Fault Diagnosis Method for Fixed-Wing UAVs Integrating Attention Mechanism and Meta-Learning DONG Qianli, ZHANG Ansi+, WU Jie, ZHAO Kaijun Computer Engineering and Applications    2025, 61 (23): 360-367.   DOI: 10.3778/j.issn.1002-8331.2409-0051 Abstract （14）      PDF（pc） （1600KB）（25）       Knowledge map    Save With the increasing application of UAVs in various fields, fault diagnosis has become crucial for ensuring their safe operation. However, traditional deep learning-based fault diagnosis methods often rely on large amounts of labeled data, leading to issues such as poor generalization performance, insufficient extraction of key features, and overfitting, especially in scenarios with small sample sizes and complex flight environments. To address these challenges, a meta-learning and effective channel attention(MLECA) fault diagnosis method is proposed. This method aims to improve the accuracy and robustness of fault diagnosis through meta-learning. Firstly, the original sensor data are preprocessed, and meta-tasks are constructed. Secondly, to effectively capture and emphasize important features, a feature encoder combining convolutional neural networks and efficient channel attention (ECA) is established. Finally, it is used as the base model, and model-agnostic meta-learning is applied to train and optimize the initialization parameters to acquire prior representational knowledge, which is then used for fixed-wing UAV fault diagnosis in unknown environments. Experimental results demonstrate that the MLECA method exhibits better overall diagnostic performance and stronger generalization capability. Reference | Related Articles | Metrics Select Instant Delivery Problem with Constrained Capacity Considering Urban Traffic Asymmetric Network WU Tengyu, XUE Huanhuan, FU Deqiang, YU Haiyan Computer Engineering and Applications    2025, 61 (23): 368-376.   DOI: 10.3778/j.issn.1002-8331.2410-0277 Abstract （17）      PDF（pc） （939KB）（18）       Knowledge map    Save With the increasing scale and range of deliveries, riders have frequent traffic accidents. The complexity of the urban traffic network and the deviation of order capacity from the platform prediction force the riders to adopt non-standard loading methods such as mounted handlebars during peak hours, which significantly increases the risk of traffic accidents. Therefore, it is essential to consider capacity-constrained pickup and delivery strategies. So a real-time delivery route optimization problem with capacity constraints considering the characteristics of urban transportation networks is proposed. Firstly, the lower bound of the problem is demonstrated. Double judgment condition (DJC), judge path and load weighted (JPL) and wait and serve (W&S) strategies for specific and general networks are designed. And worst-case scenario analysis is used to prove the competitive ratios of these strategies. Finally, through case studies and analysis of the performance of the JPL and W&S strategies under different order densities, maximum asymmetry coefficients and order capacity demand ratios, the algorithms’ effectiveness is validated. The results indicate that the JPL strategy is highly applicable and performs best in urban traffic networks with higher order density, more large-capacity orders and smaller asymmetric coefficients. The W&S strategy is more suitable for asymmetric urban transportation networks with lower order density and significant capacity demand. The conclusion of the study provides a delivery strategy considering capacity constraints in different cases, and reduces the non-standard loading demand through real-time optimization of the path, ensuring the safe delivery of riders. Reference | Related Articles | Metrics Select Stock Price Trend Prediction Based on Event-Aware LSTM Model WU Xie, ZHAO Guanying, LIU Hongzhi, NI Ziheng, SUN Tianqi Computer Engineering and Applications    2025, 61 (22): 295-303.   DOI: 10.3778/j.issn.1002-8331.2412-0437 Abstract （47）      PDF（pc） （1193KB）（71）       Knowledge map    Save Stock price trend prediction is a critical issue in securities investment. Due to the nonlinear, volatile, and noisy characteristics of stock price sequences, traditional time series forecasting models often struggle to effectively capture these dynamics. To address these challenges, this paper presents an event LSTM model that modifies the cell structure of long short-term memory (LSTM) networks, allowing for the effective integration and modeling of both non-periodic event information and periodic stock price data. Specifically, a dedicated event control gate is incorporated into the LSTM cell structure to process non-periodic event information, enabling the model to effectively capture the impact of such events on stock price trends. Additionally, an attention mechanism is utilized to highlight the contributions of historical key information, thereby reducing the influence of noisy data and achieving more accurate predictions of stock price trends. Experimental results demonstrate that the proposed model significantly outperforms existing models in terms of prediction accuracy, and exhibits exceptional performance in accounting for the effects of announcement events. Reference | Related Articles | Metrics Select Optimization Research on Multi-Path Retrieval-Augmented Generation Method for Wine Domain Knowledge YANG Wenyue, YU Qiancheng, WANG Qiming, MU Hongrui, ZHOU Chengchen Computer Engineering and Applications    2025, 61 (22): 304-319.   DOI: 10.3778/j.issn.1002-8331.2411-0470 Abstract （23）      PDF（pc） （1207KB）（25）       Knowledge map    Save In the domain of wine, designing a specialized domain knowledge Q&A system is of great significance for improving the technical skills of vineyard growers, industry workers, winemakers, and sommeliers, as well as for supporting the digital transformation of the entire industry. Given the complexity of the wine domain, which involves a wide range of specialized terminology, intricate production processes, diverse wine classifications, sensitive vintage variations, and rich winery cultures, existing AI-powered Q&A systems often fall short due to incomplete information extraction and insufficient retrieval accuracy, failing to meet the quality and accuracy requirements of responses. A method is proposed, which integrates the strengths of knowledge graphs, vector databases, and PKL file libraries to construct a wine domain knowledge base. Additionally, the multi-path retrieval-augmented generation (MPRAG) method for wine domain knowledge is explored, which enhances the precision and recall of retrieval results through parallel retrieval across different paths. Experimental results indicate that the MPRAG method improves both precision and recall by approximately 20 percentage points compared to several existing open-source RAG systems, providing support for technical assistance and talent training in the wine industry. Reference | Related Articles | Metrics Select Multi-Agent Single-Goal Collaborative Exploration in Unknown Environments with Improving MADDPG Algorithm HAN Huiyan, SHI Shuxi, KUANG Liqun, HAN Xie, XIONG Fengguang Computer Engineering and Applications    2025, 61 (22): 320-328.   DOI: 10.3778/j.issn.1002-8331.2408-0131 Abstract （23）      PDF（pc） （2698KB）（34）       Knowledge map    Save To address the inefficiency of the multi-agent deep deterministic policy gradient (MADDPG) algorithm in unknown environments, a new multi-agent deep reinforcement learning algorithm called RE-MADDPG-C is proposed. This algorithm uses residual networks (ResNet) to alleviate gradient vanishing and explosion issues, enhancing convergence speed. To tackle the convergence difficulty caused by sparse rewards in single-goal exploration in unknown environments, a multi-agent intrinsic curiosity module (ICM) is introduced. The curiosity reward serves as an intrinsic motivation for agents, providing additional exploration incentives. By designing a suitable exploration reward function, agents can accomplish single-goal tasks in unknown environments. Simulation results show that the proposed algorithm achieves faster reward improvement during training, quickly completing exploration tasks. Compared to MADDPG and other algorithms, the proposed algorithm reduces training time and achieves higher global average rewards. Reference | Related Articles | Metrics Select Crack Detection Based on Dual Branch Fusion and Multi-Scale Semantic Enhancement LI Jie, LI Huanwen, TU Jingmin, LIU Zhao, YAO Jian, LI Li Computer Engineering and Applications    2025, 61 (22): 329-338.   DOI: 10.3778/j.issn.1002-8331.2408-0152 Abstract （27）      PDF（pc） （1614KB）（49）       Knowledge map    Save Fine-grained cracks, as an early stage of road surface crack formation, can be detected and repaired to eliminate safety hazards and reduce maintenance costs in a timely manner. Fine-grained cracks not only have complex topological structures, but also have geometric characteristics of small width and variable scale. In complex road backgrounds, existing methods are prone to missed detections and have low accuracy in perceiving their width. In response to this, this paper proposes a fine-grained crack detection method for road surfaces based on dual branch selective fusion and multi-scale semantic enhancement. An enhanced self-attention mechanism and a dual branch parallel backbone network of CNN (convolutional neural network) are designed to simultaneously extract features from both local and global perspectives, enriching feature representations layer by layer. A redundancy reduction and feature selective fusion (RSF) module is proposed to achieve the learning and interaction of dual branch global and local information, enhancing the expressive power of features. A multi-scale semantic enhancement fusion strategy is adopted to enhance the model’s perception ability of fine-grained crack features through cross scale information transmission and fusion. To validate the effectiveness and reliability of the proposed method, training and testing are conducted on the CrackTree260 public dataset, and the model’s generalization performance is evaluated on the CRKWH100 dataset. The experiment shows that the proposed method achieves ODS values of 0.909 and 0.918 on two datasets, respectively, which is superior to other advanced crack detection methods. Reference | Related Articles | Metrics Select Interpretable Association Rule Defect Prediction Model Combining Counterfactuals and Multi-Objective Optimization YU Qiao, JIANG Jiaxuan, REN Siyu, ZHU Yi Computer Engineering and Applications    2025, 61 (22): 339-352.   DOI: 10.3778/j.issn.1002-8331.2501-0275 Abstract （23）      PDF（pc） （1939KB）（22）       Knowledge map    Save Software defect prediction is the key to ensure software quality. In order to improve the performance of software defect prediction, researchers have designed a variety of defect prediction models, but most of the models are less transparent in providing prediction results, which makes it difficult for developers to understand the internal logic and decision-making process of the models, and thus leads to the non-interpretability problem of the models. This problem not only limits the credibility of the models, but also hinders their application in practical development. To address this problem, this paper uses multiple association rules to combine into an interpretable multi-objective optimization model, known as MoCFR, which employs a counterfactual interpretation method for feature selection, and determines the importance score of each feature by the feature change rate of the counterfactual sample. Based on this, the model applies multi-objective optimization techniques to construct an association rule classifier, while optimizing three key metrics： classification error, average number of rules, and confidence. Experimental results on the PROMISE dataset show that MoCFR outperforms existing rule-based classification models in terms of classification error and significantly reduces the number of rules compared to similar multi-objective optimization models. Reference | Related Articles | Metrics Select Optimization of Joint Distribution Routing for Cold Chain Logistics Considering Uncertain Demand and Mixed Fleet LIN Mingxiu, CHU Liangyong, WANG Jianing, HUANG Xianting Computer Engineering and Applications    2025, 61 (22): 353-363.   DOI: 10.3778/j.issn.1002-8331.2501-0045 Abstract （34）      PDF（pc） （1340KB）（36）       Knowledge map    Save In addressing the multi-depot cold chain logistics vehicle routing problem under uncertain demand and mixed fleets, this paper comprehensively considers practical factors such as customer service levels, time windows, multi-depot settings, and joint distribution with electric and fuel vehicles. The goal is to establish a routing optimization model that minimizes the total costs, encompassing vehicle fixed costs, charging costs, fuel costs, cargo damage costs, refrigeration costs, and penalty costs. This model is deterministically transformed by using chance-constrained programming. An initial solution is generated by using the labeling method, and a hybrid improved differential evolution adaptive large neighborhood search (ALNS) algorithm is designed to solve the model. The mutation operator is designed based on a dynamic mutation strategy, and three destruction operators and three repair operators from the ALNS algorithm are introduced for search optimization. The validity of the model and algorithm is verified through practical data and case studies, and the influences of the demand variation coefficient, the number of distribution centers, and the maximum vehicle load capacity on distribution costs are analyzed. This provides a reference for enterprises to rationally allocate transportation resources and optimize decision-making in distribution schemes. Reference | Related Articles | Metrics Select Knowledge Graph-Based Problem Management of Foreign Object Debris at Aviation Product Headquarters Assembly Site QIAN Yun, GENG Xiuli Computer Engineering and Applications    2025, 61 (22): 364-372.   DOI: 10.3778/j.issn.1002-8331.2405-0129 Abstract （41）      PDF（pc） （919KB）（47）       Knowledge map    Save Effective management of foreign object debris at the headquarters assembly site of aviation products is crucial to ensure the safety and reliability of aviation products. Aiming at the problem that the management of existing foreign object debris relies on the experience of experts and is highly subjective, a method of managing foreign object debris based on ontology modeling to construct a knowledge graph is proposed. The ontology model is used to describe and organize the knowledge related to foreign object debris, including the attributes of the foreign object, the associated relationships, the causes of the foreign object debris, and so on, to construct a knowledge graph to demonstrate the concepts related to foreign object debris clearly and to provide intuitive decision-making support for foreign object debris managers. Based on the existing relationships in the knowledge graph, the relationship between the causes of the foreign object problem and the management measures is constructed. The causes of the foreign object problem are categorized by using the binary tree support vector machine, which can assist the foreign object managers finding out the root causes of the foreign object problem more quickly and taking relevant management measures. The effectiveness of the proposed method is verified through case studies, and the results show that the knowledge graph of foreign object debris based on the ontology model can provide managers with structured information on foreign object debris and provide new retrieval ideas in the process of applying the knowledge graph. Reference | Related Articles | Metrics Select Speaker Verification in Deliberately Disguised Scenarios QIN Xiaoyi, LI Ze, LIU Dong, LI Ming Computer Engineering and Applications    2025, 61 (21): 324-332.   DOI: 10.3778/j.issn.1002-8331.2408-0403 Abstract （28）      PDF（pc） （39215KB）（58）       Knowledge map    Save The challenge in the task of deliberately disguised speaker verification lies in the speaker intentionally altering their voice to become someone else and thereby concealing their identity. This task is viewed as a scenario where one person plays multiple roles, for which the CN-Movies training set and TheSound-test testing set are proposed. The CN-Movies dataset is constructed by matching characters, detecting faces, recognizing faces, lip movement recognition, and voice activity detection in Chinese movies featuring actors and voice actors. This dataset includes the original voices of actors and their corresponding voice actors, leveraging the characteristics of actors and voice actors intentionally altering their voice to portray different roles, thus facilitating the collection of multi-role data for deliberate disguise. Additionally, utilizing the feature of the program TheSound, where voice actors intentionally hide their identities to avoid being recognized, the TheSound-test is proposed as a testing set for deliberate disguise scenarios. By combining the data mined from the above fields, a siamese network model is employed, achieving significant improvements in speaker verification performance on both the VoxMovies test set and TheSound-test set. Reference | Related Articles | Metrics Select Research on Digital Twin System of Biped Robots and Reality-Virtual Communication Strategy ZHANG Deyu, LIU Siyu, YUAN Changshun Computer Engineering and Applications    2025, 61 (21): 333-341.   DOI: 10.3778/j.issn.1002-8331.2407-0352 Abstract （25）      PDF（pc） （3013KB）（28）       Knowledge map    Save Based on OpenGL tools and the user datagram protocol (UDP), this paper proposes a digital twin system for biped robots with multi-type sensor data integration capabilities and its reality-virtual communication strategy. This addresses the dependency of domestic biped robots on foreign robot operating systems (ROS) at the digital twin level. Additionally, to tackle the issue of inconsistency between the actual physical characteristics and the state of the digital twin model caused by joint errors in biped robots, an effective error control method is proposed. Experimental results show that the system has real-time interaction capabilities, with data delay reduced by 46% compared to the widely used ROS framework, while the error control method reduces the average spatial error of the robot’s limbs in the virtual environment by 51%. This paper provides a fast, lightweight implementation solution for the digital twin systems of future domestic robots. Reference | Related Articles | Metrics Select Graph Embedded Deep Reinforcement Learning Strategy for Flexible Job Shop Scheduling CHEN Mingtong, ZHANG Jianxin, HOU Shujun Computer Engineering and Applications    2025, 61 (21): 342-350.   DOI: 10.3778/j.issn.1002-8331.2504-0299 Abstract （27）      PDF（pc） （1670KB）（38）       Knowledge map    Save The flexible job shop scheduling problem (FJSP), characterized by its NP-hard nature, complex resource constraints, and strong coupling of operations, presents significant challenges in large-scale instances in terms of both solution quality and computational efficiency. Traditional heuristic methods often suffer from limited scalability and poor generalization. To address these limitations, this paper proposes the SMG-DRL scheduling framework, which integrates graph neural networks (GNNs) with deep reinforcement learning (DRL), and incorporates three key designs: state load awareness, mask-guided action selection, and graph structural enhancement. Specifically, SMG-DRL embeds both global and local state features via load-aware mechanisms, improves action selection efficiency through a masking strategy, and utilizes sparse graph modeling combined with graph isomorphism networks (GIN) to extract structural features and reduce graph density. Experimental results on the Brandimarte benchmark dataset demonstrate that SMG-DRL achieves approximately 42% reduction in average relative error (Gap) compared to three classic dispatching rules (SPT, MWKR, and LWKR), with the latter yielding an average Gap of 35.55%, while the proposed framework achieves 20.55% on the 10×5 and 20×10 training instances. Moreover, compared with the DRL-based methods proposed by Zhang and Lei, SMG-DRL exhibits superior performance in both Gap levels and solution stability, reflecting stronger convergence and robustness. In terms of computational efficiency, SMG-DRL reduces the average solving time from 978 seconds (OR-Tools) to approximately 3 seconds, achieving a nearly 300-fold speed-up. Furthermore, on large-scale instances (30×10 and 40×10), the framework maintains low Gap values between 6.23% and 8.76%, showcasing its excellent scalability and generalization capability. Reference | Related Articles | Metrics Select Fusion of Online Label Smoothing Strategy and Relational Network for TCM Syndrome Classification Model LIAO Ming, DU Jianqiang, LUO Jigen, HUANG Qiang, HE Jia, FAN Yue Computer Engineering and Applications    2025, 61 (21): 351-360.   DOI: 10.3778/j.issn.1002-8331.2407-0408 Abstract （34）      PDF（pc） （952KB）（34）       Knowledge map    Save In the research on intelligent classification of TCM syndromes, due to class imbalance and the scarcity of high-quality manually labeled samples, the ability of model to learn from few-sample labels is inadequate, resulting in unsatisfactory overall classification performance. To address these issues, an online label smoothing strategy integrated with relational networks for TCM syndrome classification models (online label smoothing for relational networks based on pre-trained language models, PLM-SNet) is proposed. This model encodes the input case text using pre-trained language models to obtain feature representations of the input samples. It cascades the feature information through a relational network using the support set and query set of the samples to obtain the relevance scores of the samples in the query set. The training loss of the online label supervision model is used to update and optimize the soft labels of the categories in real-time, resulting in the final category scores. Experimental results on the TCM public dataset TCM-SD and the self-constructed TCM asthma dataset J-SD show that compared to the optimal baseline model, PLM-SNet improves the Macro-F1 and G-Mean of TCM syndrome classification by 3.47, 2.48, 3.06, and 2.58 percentage points, respectively. The experimental results verify the scientific validity and effectiveness of the model in the class imbalance TCM syndrome classification task. Reference | Related Articles | Metrics Select Multidimensional Spatiotemporal Interaction Network for Agricultural Machinery Trajectory Operation Mode Identification LUO Yuchuan, ZHAI Weixin Computer Engineering and Applications    2025, 61 (20): 341-357.   DOI: 10.3778/j.issn.1002-8331.2505-0298 Abstract （52）      PDF（pc） （6372KB）（58）       Knowledge map    Save Agricultural machinery trajectory operation mode identification is a multivariate time series classification (MTSC) task, which aims to extract the spatiotemporal features embedded in agricultural machine trajectory data to identify the agricultural machine trajectory operation mode and assign the corresponding semantic labels to each trajectory point. Aiming at the problems of insufficient ability to capture spatiotemporal information of trajectories and poor identification accuracy of existing methods, multidimensional spatiotemporal interaction network (MSINet) is proposed to identify the operation mode of agricultural machine trajectories. Firstly, a multidimensional information interaction (MII) module is proposed, which integrates graph convolution and self-attention mechanisms to capture local associations and global dependencies between trajectory points, and also utilizes a bidirectional perception mechanism to achieve complementary information between channel and spatial dimensions. Subsequently, the multipath feature extraction (MFE) module is designed to effectively extract multi-scale temporal features of trajectory data using convolutional paths with different dilation rates. Finally, a semantic focus (SF) module is developed to efficiently capture the key information in the trajectory data. To verify the effectiveness of the proposed method, experiments have conducted on the trajectory dataset provided by the Key Laboratory of Agricultural Machinery Monitoring and Big Data Applications Ministry of Agriculture and Rural Affairs. The experimental results show that the accuracy of MSINet on the rice harvester and wheat harvester trajectory datasets is 91.62% and 91.34%, respectively, and the F1 score is 91.57% and 86.98%, respectively. Compared with the current best-performing model generative adversarial network-bidirectional long short-term memory network(GAN-BiLSTM), the F1 score has increased by 5.57 and 3.72 percentage points, respectively. Reference | Related Articles | Metrics Select Screw Status Detection of Subway Gearbox Based on Improved YOLOv5s CHENG Guangyao, ZHANG Wenqiang, WANG Zhicheng, HUA Lujie, YAN Shichang, ZHANG Tao Computer Engineering and Applications    2025, 61 (20): 358-367.   DOI: 10.3778/j.issn.1002-8331.2406-0111 Abstract （93）      PDF（pc） （1213KB）（58）       Knowledge map    Save With the rapid development of urban rail transit, the safety and efficiency of subway maintenance have become particularly important. The condition of gearbox screws under the train is directly related to the safe operation of urban rail vehicles. However, traditional detection methods based on YOLOv5s have significant deficiencies in real-time performance. To address these issues, this paper proposes an improved YOLOv5s-based algorithm for detecting the condition of gearbox screws. Firstly, to ensure the real-time performance of the model, a group shuffle convolution module is used, which significantly reduces the calculation of the model. Secondly, to address the issue that traditional spatial pyramid pooling technology fails to fully consider the differences between features, a spatial pyramid pooling technique with feature adaptive fusion is proposed, enhancing the model’s performance. Next, to improve the representative ability of a single convolution, a new feature extraction unit is employed, which can acquire more feature information while ensuring lightweight design. Finally, to enhance the model’s ability to detect small screw targets, the Focal Loss function is used, improving detection accuracy. Experimental results show that the improved model increases mAP accuracy by 2.2 percen-tage points compared to the original YOLOv5s algorithm, while the parameters and computational load are reduced by 2.9[×]106 and 8.1 GFLOPs respectively, and the detection speed decreases to 25 ms, with a 22 percent improvement, demonstrating the effectiveness of this method. Reference | Related Articles | Metrics Select Insulator Defect Detection Based on Multi-Scale Fusion and Contextual Enhancement ZENG Yezhan, CHEN Tianhang, DENG Qian, PENG Xinyao, OUYANG Hongbo, ZHONG Chunliang Computer Engineering and Applications    2025, 61 (20): 368-378.   DOI: 10.3778/j.issn.1002-8331.2406-0331 Abstract （56）      PDF（pc） （2880KB）（48）       Knowledge map    Save The detection of insulator defects plays a pivotal role in ensuring the steady and reliable operation of power grids. However, due to the complex background and the varying scales, it is very difficult to effectively detect the insulator defect. To address this problem, a YOLOv8 model based on multi-scale fusion and contextual enhancement (MFCE-YOLOv8) is proposed. Firstly, a multi-scale information aggregation attention (MIAA) is developed to increase the capability of the detection of small-scale defects while reducing background interference. Next, to minimize the loss of feature information in the deep network, a context feature learning module (CFLM) is constructed based on global and local information. Finally, a cross-layer feature fusion module (CFFM) is proposed to fully explore information communication between different layers and reduce semantic conflicts. Experimental results show that MFCE-YOLOv8 achieves an overall mean average precision (mAP) of 92.3% on the insulator defect dataset, and obtains the AP of flashover and damage defect detection of 86.0% and 91.4%, respectively, which outperforms some existing methods. Reference | Related Articles | Metrics Select Joint Scheduling of Berths and Quay Cranes Considering Tides and Uncertainty in Ship Arrival Times ZHANG Jiawen, CHU Liangyong Computer Engineering and Applications    2025, 61 (20): 379-390.   DOI: 10.3778/j.issn.1002-8331.2407-0076 Abstract （45）      PDF（pc） （2153KB）（28）       Knowledge map    Save Berths and quay cranes are critical resources for ports, and their scheduling directly impacts port operational efficiency. This paper focuses on the joint scheduling of berths and quay cranes in ports, addressing issues such as the significant impact of tides and limited scheduling resources. Considering uncertainties in ship arrival times, ship berthing preferences, time-varying numbers of quay cranes, and the constraint that quay cranes cannot cross each other, the paper aims to minimize the sum of ship waiting time costs, delayed departure costs, fuel consumption during the port stay, and quay crane handling costs. This paper constructs a mixed-integer programming model. Given the complexity of the problem, it employs an adaptive large neighborhood search algorithm to solve the model, and designs various destruction and repair operators tailored to the problem characteristics to improve the solution accuracy and efficiency of the algorithm. Finally, the effectiveness and accuracy of the algorithm are validated through comparison with CPLEX. Its efficiency is confirmed through comparison with various other algorithms. Sensitivity analysis of key parameters provides recommendations for joint scheduling of berths and quay cranes, offering practical references for port operations. Reference | Related Articles | Metrics Select Multi-Vehicle Cooperative Control at Intersection：Recurrent Graph Attention Reinforcement Learning YANG Weida, WU Zhizhou, LIANG Yunyi Computer Engineering and Applications    2025, 61 (19): 282-291.   DOI: 10.3778/j.issn.1002-8331.2406-0105 Abstract （39）      PDF（pc） （1927KB）（30）       Knowledge map    Save There will be mixed traffic consisting of connected automated vehicles (CAVs) and human-drive vehicles (HVs) at future unsignalized intersections for long periods of time. CAV can only make decisions based on neighborhood information of mixed traffic due to CAV’s partial observation. The process of multi-vehicle cooperative control at unsignalized intersections is modeled as a decentralized partially observable Markov decision process. This study proposes a centralized training and decentralized execution (CTDE) framework based on soft actor-critic (SAC). The relationship between vehicles is modeled into a graph, and multiple attention layers are used as the convolution kernel of actor network and critic network to infer the neighbor graph features of vehicles. Gated recurrent unit (GRU) is developed to keep the long-term memory of neighbor dynamic graph features and avoid information for-getting caused by the changes of neighborhood information during vehicle movement. Simulation results show that the collision rate of this algorithm is 0. The average vehicle speed of one-way 1×1 intersection, multi-way 1×1 intersection, and multi-way 2×2 intersection network is increased by 10.51%, 4.64%, and 10.24% compared with the state-of-the-art multi-vehicles cooperative control algorithms, respectively. Reference | Related Articles | Metrics Select Research on Construction of Knowledge Graph of Textile Surface Defect JIANG Xiaoheng, LYU Pengshuai, LIU Yun, LU Yang, ZHANG Kunli, XU Mingliang Computer Engineering and Applications    2025, 61 (19): 292-301.   DOI: 10.3778/j.issn.1002-8331.2406-0087 Abstract （36）      PDF（pc） （1235KB）（21）       Knowledge map    Save Aiming at the problems of strong subjectivity and dispersed structure in the field of textile surface defects, a knowledge graph construction method for textile surface defects is studied.The entity types and relationship types of textile surface defects are defined, and a total of 7 entity categories and 26 relationship categories are divided. Based on this, a textile surface defect knowledge graph data set is constructed. At the same time, the entity extraction model based on dictionary feature enhancement and the dual-branch relationship extraction model based on residual attention enhancement are combined to automatically extract entity relationships and construct a knowledge graph of textile surface defects. Experimental results show that the F1 value of this model reaches 91% in both named entity recognition and relationship extraction tasks in the field of textile surface defects. In addition, the Neo4j database is used to store and visualize the knowledge graph, and provides technical support for downstream applications of the textile surface defect knowledge graph, such as intelligent question answering. Reference | Related Articles | Metrics Select Improved A* Algorithm for UAV Path Planning in Urban Low-Altitude Logistics ZHU Wenjie, LI Wei, WANG Ziyan Computer Engineering and Applications    2025, 61 (19): 302-310.   DOI: 10.3778/j.issn.1002-8331.2503-0300 Abstract （64）      PDF（pc） （1740KB）（54）       Knowledge map    Save To address the efficiency and safety challenges of UAVs in low-altitude logistics distribution within urban complex building clusters, an improved A* algorithm is proposed. First, the heuristic function is refined to resolve the inefficiency caused by redundant node searches in traditional A* algorithms. The three-dimensional vector cross product is introduced to eliminate decision-making uncertainties during the search process, thereby constraining the expansion scope of neighboring nodes. Second, the cost function is optimized by dynamically adjusting the heuristic weight coefficient based on the relative distance between nodes and the target, achieving a balance between search speed and path quality for safe and efficient planning. The algorithm is validated through grid-based 3D map simulations, demonstrating a 57.40% reduction in runtime, a 54.75% decrease in traversed grid nodes, and a 78.16% decline in path inflection points compared to the traditional A* algorithm. Furthermore, a UAV control platform is implemented using the ROS (robot operating system), and real-world experiments confirm the enhanced accuracy and operational efficiency of the improved algorithm in practical applications. Reference | Related Articles | Metrics Select Improving YOLOv8 Method for Coal Mine Conveyor Belt Foreign Body Detection YANG Di, ZHAO Peipei, SUN Aoran, ZHANG Junyi, XIAO Tao Computer Engineering and Applications    2025, 61 (19): 311-319.   DOI: 10.3778/j.issn.1002-8331.2410-0425 Abstract （47）      PDF（pc） （2145KB）（45）       Knowledge map    Save At present, there are some problems such as poor clarity in dark environment, image noise, high complexity of existing detection algorithm model, missing and wrong detection, and insufficient accuracy. Based on the above problems, an improved YOLOv8 model is proposed. Firstly, lightweight ShuffleNetv2 is used as the header network to reduce the number of model parameters, and multi-scale composite attention module (MSCAM) is constructed to enhance the feature extraction capability to reduce the miss and error detection rates. Secondly, in view of the low accuracy of existing models for long foreign bodies, the idea of dynamic deformable convolution (DDConv) is used in C2f module to make it easier to extract the structural features of long foreign bodies. Finally, the new SIoU loss function with angle loss is used to improve the training ability and inference performance of the model. The MTBID dataset is verified. The experimental results show that: mAP@0.5 and mAP@0.5：0.95 of the improved model can reach 0.893, 0.663, and the number of parameters is reduced by 27.6% compared with YOLOv8n. Reference | Related Articles | Metrics Select Path Planning of Street Battle Search and Rescue Based on Improved Dung Beetle Optimization Algorithm LEI Fuqiang, CHENG Zheng, XUE Zhengyu, GUAN Peng Computer Engineering and Applications    2025, 61 (19): 320-335.   DOI: 10.3778/j.issn.1002-8331.2409-0301 Abstract （52）      PDF（pc） （7650KB）（52）       Knowledge map    Save To address the issues of global search stability and susceptibility to local optima in traditional dung beetle optimization (DBO) algorithm for search and rescue path planning in urban warfare environments, this paper proposes an improved dung beetle optimization (IDBO) algorithm based on a hybrid strategy to enhance path planning efficiency and reliability. The IDBO algorithm introduces refractive reverse learning and elite selection strategies to increase population diversity and global search capability. In the rolling phase, it combines the osprey optimization algorithm (OOA) with the optimal solution to overcome the reliance on the worst individuals, enhancing global search capability in complex terrains. In the breeding phase, a dynamic selection mechanism and adaptive [t]-distribution strategy are employed to balance global exploration and local exploitation, meeting the dual requirements for accuracy and speed in search and rescue tasks. In the foraging phase, the Jacobi curve is incorporated to strengthen the algorithm’s ability to escape local optima, enabling it to effectively handle various uncertainties in urban warfare environments. Performance tests on the CEC2005 function set demonstrate that IDBO outperforms the DBO algorithm in global search capability and convergence accuracy. In path planning experiments within a simulated urban warfare search and rescue environment, in the static environment, the IDBO algorithm achieves shortest paths of 27.841 and 57.256 in simple and complex grid maps respectively, representing reductions of 2.57% and 15.35% compared to the DBO algorithm. In a dynamic environment, the shortest paths are 29.213 and 59.367, which are 3.85% and 14.37% shorter than those generated by the DBO algorithm, further validating its effectiveness and stability in urban warfare search and rescue path planning. Reference | Related Articles | Metrics Select Train Mileage Positioning Method Based on Monocular Vision Semantic Map Construction JIANG Xinlan, WANG Shengchun, SHEN Yanlong Computer Engineering and Applications    2025, 61 (19): 336-347.   DOI: 10.3778/j.issn.1002-8331.2501-0216 Abstract （38）      PDF（pc） （4365KB）（18）       Knowledge map    Save To address the limitations of traditional mileage positioning methods used in railway inspection trains, this paper integrates semantic information into 3D reconstruction technology and proposes a monocular vision-based semantic map construction approach using prior knowledge. The method consists of modules for object detection, object tracking, monocular SLAM, and point cloud statistical processing. It optimizes key challenges in 3D reconstruction of railway scenes by mitigating the scale drift issue through locating stable semantic targets, reducing feature point matching errors caused by the uniformity of railway scene patterns and excessive train speed by setting adaptive thresholds and selecting feature points with good 3D reconstruction consistency, and enhancing the algorithm’s computational efficiency by filtering the results of the intrinsic matrix’s singular value decomposition. Experimental results demonstrate that the proposed method is highly feasible, easy to apply, and provides high-precision railway mileage positioning. Compared to mileage positioning based solely on wheel encoders, the method can limit the positioning error of the detection system to approximately 1 meter, reducing the maximum error from 11.3 meters to 1.7 meters, thereby achieving precise mileage positioning for inspection trains. Reference | Related Articles | Metrics Select Research on Detection of Rail Surface Defects by Combining Spatial Domain and Frequency Domain Processing Techniques JI Menghao, WANG Xin Computer Engineering and Applications    2025, 61 (19): 348-358.   DOI: 10.3778/j.issn.1002-8331.2405-0236 Abstract （53）      PDF（pc） （3075KB）（32）       Knowledge map    Save Aiming at the problem of image blur and distortion caused by camera shake, which affects the detection performance in the task of rail surface defect detection, this paper proposes an image enhancement algorithm that combines spatial domain and frequency domain processing. Additionally, an improved YOLOv8s model is presented for rail surface defect detection. The purpose is to enhance image quality to improve the detection accuracy of the detection network. The images are deblurred and restored using image restoration network, with deep learning utilized to reduce the detection region. An adaptive gamma transformation, which integrates global and local spatial information, along with a frequency domain filtering algorithm improved via Laplacian pyramids, is proposed to enhance image brightness, contrast, and detail information. Furthermore, the enhanced images are trained using the YOLOv8s algorithm, with enhancements in the feature fusion network, the integration of attention modules, and the design of a detection head based on parameter sharing to reduce network parameters and increase detection precision. Experimental results demonstrate that compared to datasets not using the image enhancement algorithm, the enhanced dataset achieves better detection accuracy; when trained on the enhanced dataset, the improved detection network shows a significant improvement in mean average precision across all targets, large targets, and small targets. Reference | Related Articles | Metrics Select Identification of Hose Whipping Phenomenon in Aerial Refueling Based on Real-Time Image Segmentation and Trajectory Prediction ZENG Bohan, QIE Rongkai, WANG Xuan, ZHANG Zhaoxiang, XU Yuelei, KANG Mengte Computer Engineering and Applications    2025, 61 (19): 359-370.   DOI: 10.3778/j.issn.1002-8331.2405-0418 Abstract （35）      PDF（pc） （8668KB）（29）       Knowledge map    Save Hose whipping phenomenon (HWP) poses a threat to flight safety during aerial refueling docking. Existing research focuses on physical modeling, while there is a research gap in the visual domain. After analyzing the spatiotemporal characteristics of HWP, a vision-based HWP emergency identification algorithm is proposed, which includes three steps：hose segmentation, trajectory prediction, and multi-criteria discrimination. A warning system is developed to enhance the safety of aerial refueling missions. To balance segmentation speed and accuracy, a real-time segmentation network is proposed. By introducing an adaptive regions self-attention mechanism, the network can effectively and accurately segment the hose area, achieving  mIoU of 82.6%, which surpasses most existing real-time segmentation algorithms. A relative distance loss function is designed to train long short-term memory network for predicting the hose trajectory. The effectiveness of the hose segmentation and trajectory prediction algorithms is validated in different environments. Hose whipping phenomenon identification experiments based on monocular vision guidance are conducted in a simulation environment. Multiple criteria are used to achieve discrimination and warning of HWP emergencies, achieving an 81.6% success rate under clear weather conditions. Reference | Related Articles | Metrics Select Research on 6D Robotic Arm Grasping Method Based on Improved DDPG ZHANG Sheng, SHEN Jie, CAO Kai, DAI Huishuai, LI Tao Computer Engineering and Applications    2025, 61 (18): 317-325.   DOI: 10.3778/j.issn.1002-8331.2410-0186 Abstract （61）      PDF（pc） （2758KB）（72）       Knowledge map    Save In current 6D robotic grasping tasks based on deep reinforcement learning, suboptimal grasping poses often lead to insufficient grasping success rates and robustness. To address this issue, an improved DDPG algorithm incorporating a pose evaluation mechanism is proposed. Building upon the DDPG framework,the algorithm introduces a grasp evaluation network to quantitatively assess the grasping poses of the robotic manipulator. Based on the evaluation scores, multi-level reward values are assigned to the grasping actions, enabling the assessment of pose quality and guiding the DDPG to optimize grasping poses. Experiments conducted in both simulation and physical environments demonstrate that the proposed method effectively improves the grasping poses of the robotic manipulator and enhances the grasping success rate. Furthermore, the method exhibits strong adaptability to real-world scenarios,significantly improving the generalization capability and robustness of the robotic manipulator. Reference | Related Articles | Metrics Select Research on Global Path Planning for Unmanned Vehicles Based on Improved RRT* Algorithm DAN Yuanhong, HUANG Binbin, FENG Guangxu Computer Engineering and Applications    2025, 61 (18): 326-335.   DOI: 10.3778/j.issn.1002-8331.2501-0117 Abstract （66）      PDF（pc） （1307KB）（62）       Knowledge map    Save An improved RRT* algorithm is proposed to address the issues of low node expansion efficiency, large search space, and path curvature in the global path planning of autonomous vehicles. In this approach, an adaptive bias sampling strategy is employed to adjust the sampling points towards the goal, thereby improving the expansion quality. During the expansion phase, multiple candidate nodes are selected, with dynamic adjustment of step sizes based on actual and potential costs, enhancing the algorithm’s adaptability to different environments. After the initial path is generated, heuristic cost-based sampling is used at the nodes with the largest heuristic cost to accelerate path convergence. Finally, bidirectional path optimization based on line-of-sight checking and an improved B-spline interpolation method are applied to post-process the path, improving its smoothness. Simulation results demonstrate that the proposed algorithm significantly outperforms other algorithms of the same type in terms of path planning efficiency, path cost, and smoothness, providing a reliable solution for quickly obtaining a collision-free and smooth global optimal path for autonomous vehicles. Reference | Related Articles | Metrics Select Location-Routing Optimization of Multi-Cycle Precooling Facilities for Agricultural Products Under Uncertain Demand WU Nuan, TAN Liqi, DU Jian Computer Engineering and Applications    2025, 61 (18): 336-346.   DOI: 10.3778/j.issn.1002-8331.2407-0023 Abstract （39）      PDF（pc） （1071KB）（26）       Knowledge map    Save In response to the problems of high cost and poor flexibility of traditional fixed cold storage, this paper proposes a coordinated pre-cooling mechanism that integrated pre-cooling vehicles and mobile cold storage, taking into account the characteristics of periodic changes and uncertain demand of agricultural products. A multi-cycle agricultural product pre-cooling facility site selection path optimization model is constructed with the objective of minimizing system cost, considering variables such as pre-cooling mode, yard location, relevant vehicle types and quantities, and paths. And the k-means clustering algorithm is used to complete the location problem and customer group division of the parking lot, and a hybrid adaptive large-scale neighborhood search genetic algorithm (HALNS-GA) is designed to complete routing optimization. Finally, the effectiveness of the model and algorithm is verified through numerical examples. By comparing the costs under different pre-cooling mechanisms and analyzing the sensitivity of important parameters, the economic feasibility of the pre-cooling mechanism has been verified. This study can provide reference for the layout planning and routing of pre-cooling services in rural areas of China. Reference | Related Articles | Metrics Select Dynamic Cold Chain Delivery Route Optimization for Minimizing Expected Costs YU Haiyan, LIN Yuting, WU Tengyu Computer Engineering and Applications    2025, 61 (18): 347-357.   DOI: 10.3778/j.issn.1002-8331.2412-0047 Abstract （47）      PDF（pc） （3739KB）（45）       Knowledge map    Save To address the challenges of scheduling difficulties and high delivery costs caused by dynamically emerging cold chain delivery demands, this study proposes a proactive delivery model that integrates demand forecasting. To balance the spatial waste caused by early handling of dynamic demands with the additional vehicle dispatching costs incurred by real-time demand responses, the model leverages the probability of dynamic demand occurrence to plan delivery routes in advance, aiming to minimize the expected delivery cost. The first step involves developing a dynamic demand forecasting model to estimate the probability of demand occurrence. A cold chain delivery path optimization model for static demand is then constructed. Based on this static model, a dynamic cold chain delivery routing optimization model is constructed, considering different scenarios of dynamic demand occurrence and joint delivery with static demands, with the objective of minimizing expected costs. A two-stage solution algorithm is designed. For static demands, an improved genetic algorithm is proposed to generate an initial pre-optimized route. For dynamic demands, an expected route adjustment algorithm is designed based on the pre-optimized static demand route to plan a delivery scheme that minimizes expected costs before dispatch. Finally, the proposed delivery model is validated through a case study of a pharmaceutical cold chain logistics company in Chongqing, China, along with tests on datasets of varying scales. Comparative analyses are conducted against models that handle all dynamic demands in advance and models that respond to demands in real-time. The results demonstrate the effectiveness of the proposed model and algorithm, showing that this delivery model can more efficiently plan dynamic cold chain delivery routes and significantly reduce total delivery costs, providing valuable insights for optimizing cold chain delivery routing under dynamic demand conditions. Reference | Related Articles | Metrics Select Research on Optimization of Electric Truck Distribution Route Considering Influence of Time-of-Sale Price ZHU Changzheng, CHEN Yang, LU Liang Computer Engineering and Applications    2025, 61 (18): 358-367.   DOI: 10.3778/j.issn.1002-8331.2405-0252 Abstract （60）      PDF（pc） （884KB）（33）       Knowledge map    Save The use of electric trucks to replace traditional fuel vehicles for urban distribution has become a trend in the development of the logistics industry. When using electric trucks, the impact of electricity prices at different times on distribution costs is significant. A mathematical model is established to study the distribution route problem of electric trucks under the factor of time of use electricity price, taking into account different electricity prices, customer time constraints, and constraints on electric truck electricity quantity, with the goal of minimizing the total distribution cost by making the departure time of each vehicle different. The simulated annealing algorithm is improved based on the model and validated it with the Solomon dataset. The example results indicate that the changes in electricity prices of different magnitudes during the distribution process have a significant impact on the target value. Finally, the charging strategy under different electricity prices is obtained: selecting the 30%~70% range under high electricity prices and the 70%~100% range under low electricity prices. It has reference significance for logistics and distribution enterprises. Reference | Related Articles | Metrics Select Dual-Stream Relation-Aware Attention Guided Domain Adaptation for Cancer Diagnosis SHI Hang, WU Yawen, ZHANG Daoqiang, SHAO Wei Computer Engineering and Applications    2025, 61 (18): 368-376.   DOI: 10.3778/j.issn.1002-8331.2405-0313 Abstract （52）      PDF（pc） （3968KB）（64）       Knowledge map    Save Recently, with the rapid development of digital photography technology and deep learning algorithms, the use of deep learning algorithms for processing and analyzing digital pathology images has become a hot topic. However, traditional fully supervised methods require a quantity of annotated data, which is labor-intensive and costly due to the diversity and heterogeneity of cancer subtypes. To reduce the cost of annotation, researchers have started exploring how to assist in the diagnosis of cancer subtypes with limited or no labeled data by transferring the knowledge learned from labeled cancer data. Nevertheless, existing transfer learning methods often overlook the similar spatial interaction relationships between tumors and tumor-infiltrating lymphocytes (TILs). In addition, the direct alignment of feature spaces between source and target domains may lead to negative transfer phenomena. Based on this, a dual-stream relation-aware attention （DRA） guided domain adaptation for cancer diagnosis is proposed. Specifically, DRA model consists of two modules to learn the transferable components: the bag-level transferable attention module (BTA) and the patch-level transferable attention module (PTA). The former BTA learns the intra-domain and inter-domain spatial interaction relationship by the self and cross-attention mechanism, respectively. Meanwhile, the latter PTA is introduced to identify the candidate patches for the spatial interaction relationship transfer. Proposed method is evaluated on two WSI datasets and the superior performance demonstrates the effectiveness of the DRA method in cross-domain cancer diagnosis. Reference | Related Articles | Metrics Select Heuristic Deep Reinforcement Learning Algorithm for Solving Online 3D Bin Packing Problem ZHANG Changyong, YAO Kaichao, ZHANG Yuhao Computer Engineering and Applications    2025, 61 (17): 329-336.   DOI: 10.3778/j.issn.1002-8331.2407-0069 Abstract （72）      PDF（pc） （1842KB）（41）       Knowledge map    Save Cargo loading is a key part of the logistics transportation process, which belongs to the NP-Hard problem. In order to solve the real-time problem of “real-time palletizing” in the field of intelligent logistics, a candidate heuristic deep reinforcement learning algorithm for online 3D case loading is proposed. The palletizing process of online 3D bin packing is expressed as a Markov decision process, and the reinforcement learning elements are designed to satisfy seven practical constraints. A candidate cache is set on the basis of the deep reinforcement learning algorithm. Candidate solutions are generated according to the heuristic experience, from which the algorithm filters the optimal solutions for training, and finally outputs the optimal action after evaluation by the dueling network. Experimental results show that the algorithm has a space utilization of 85.3%, a 25% improvement in convergence speed, and an average of 15 ms faster decision time, which effectively solves the problem of initial exploration difficulty of the intelligent body caused by facing the large-scale growth of the action space, and improves the efficiency and practicability of the algorithm, which is more suitable for the actual on-line crating scenarios. Reference | Related Articles | Metrics Select Research on Cheating Suppression System Based on Virtual Reality Technology PAN Di, WANG Xiaochuan, LI Haisheng Computer Engineering and Applications    2025, 61 (17): 337-343.   DOI: 10.3778/j.issn.1002-8331.2405-0017 Abstract （54）      PDF（pc） （1415KB）（47）       Knowledge map    Save Addressing the prevalent issue of exam cheating among college students and the inadequacy of conventional disciplinary and educational methods in deterring cheating motivations, this paper presents a simulation system based on virtual reality (VR) technology and embodied cognition theory, specifically designed for exam cheating behavior suppression. Employing the Unity engine, the system combines virtual examination environment construction, virtual character action simulation, virtual perspective depth of field effects, and a cheating violation detection mechanism. It thereby provides students with experiential understanding and perception of cheating behaviors from different roles. The experimental outcomes substantiate that, after the utilization of this system with virtual examination scenarios from both the perspectives of cheating students and invigilators, the mean score for participants’ attitude against cheating is escalated by 2.707%. Moreover, the average score for commitment towards abstaining from cheating is improved by 2.233%, and the mean score for response against cheating behavior is increased by 4.481%. These findings collectively indicate the effectiveness of this system in curbing the cheating inclination among participants. Reference | Related Articles | Metrics Select Multi-Intersection Traffic Signal Cooperative Control Method Based on Multi-Agent Sequential Decision Making WANG Zhiwen, LU Yumei, ZHANG Haipeng, PANG Yuli Computer Engineering and Applications    2025, 61 (17): 344-354.   DOI: 10.3778/j.issn.1002-8331.2405-0153 Abstract （70）      PDF（pc） （2641KB）（54）       Knowledge map    Save Deep reinforcement learning can use the advantages of large sequence models to solve the problem of multi-intersection traffic signal cooperative control, and a multi-agent sequential decision-making method for coordinated control of multi-intersection traffic signals is proposed. Firstly, according to the multi-agent dominance decomposition theorem, the multi-intersection traffic signal control is modeled as a sequence problem by using the characteristics of the sequence models, and the real-time multi-intersection traffic signal control is transformed into a multi-agent sequence decision-making problem, which makes full use of the amazing relationship between the multi-agent reinforcement learning decision-making process and the sequence model prediction. Then, the small-sample Transformer sequence model is used to learn the optimal control strategy of each agent online to realize the cooperative control of traffic signals at multiple intersections, which solves the problem that it is difficult to cover all the complexity of multi-agent interaction in the training mode of centralized training and decentralized execution, and the optimal joint value function is more complex to solve with the increasing number of agents. The experimental results show that the proposed method can significantly improve the performance of the traffic signal control algorithm and reduce the complexity of its implementation. Reference | Related Articles | Metrics page Page 1 of 4 Total 136 records First page Prev page Next page Last page