最新刊期

    23 2 2024
    • ZHOU Zeqing,YU Yuecheng,GU Zhouyu
      Vol. 23, Issue 2, Pages: 1-8(2024) DOI: 10.11907/rjdk.222466
      摘要:In order to solve the problem that the algorithm has low detection accuracy for small targets due to insufficient number of small target samples in the dataset and unbalanced distribution. Based on the multi-scale target detection model YOLOv4, YOLOv4-Balance was proposed which combines dynamic loss feedback and data augmentation. First, in order to balance the distribution of small target samples and enrich the quality of small target samples in the data set, a data enhancement algorithm U-Mix based on image combination and stitching is proposed. Secondly, based on the Loss feedback during model training iterations, a multi-scale model training algorithm DLF(Dynamic loss feedback) using dynamic Loss feedback is proposed to improve the contribution of the small target samples to the model during the training process. The experimental results show that in the MS COCO dataset, compared with the baseline model YOLOv4, the average accuracy of YOLOv4-Balance is improved by 2.1% and the detection accuracy for small target samples is improved by 2.8%. The algorithm in this paper will not introduce additional computational overhead, and the model converges quickly, which is conducive to efficient training.  
      关键词:object detection;sample balance;data enhancement;feedback drive;multiscale   
      195
      |
      47
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 44816222 false
      发布时间:2024-02-22
    • LIAO Xuechao,CHENG Yiqun
      Vol. 23, Issue 2, Pages: 9-16(2024) DOI: 10.11907/rjdk.231073
      摘要:In order to realize efficient scheduling of wind farm energy use management and fully extract the potential relationship between spatial and temporal characteristics of multiple sites, a multi-site short-term wind power spatio-temporal combination prediction model based on dynamic graph convolution and graph attention is proposed. Firstly, graph convolution is used to realize neighbor aggregation of temporal features among multiple sites, and graph attention mechanism is used to enhance its ability to extract spatial features. At the same time, in view of the problem that the traditional model cannot handle the real-time changes of the graph node correlation, firstly, the adjacency matrix is dynamically constructed according to the correlation coefficient and distance between the sites during the graph convolution process; secondly, the gated cycle unit is used to process the context information of the output of the dynamic graph convolution; finally, the wind power prediction is completed. The experimental results show that the proposed combined model is optimal in terms of prediction accuracy, stability and multi-step prediction performance.  
      关键词:short-term wind power forecast;dynamic correlation;graph convolution neural network;attentional mechanism;gated recurrent unit   
      69
      |
      12
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221435 false
      发布时间:2024-02-22
    • LIU Jiawei,ZHENG Junhong,HE Lili
      Vol. 23, Issue 2, Pages: 17-24(2024) DOI: 10.11907/rjdk.231114
      摘要:A multi feature fusion ALBERT-SA-BIGRU model is proposed to address the issue of insufficient consideration of customer emotional characteristics in enterprise marketing activities, resulting in unsatisfactory marketing outcomes. Firstly, based on the barrage data of enterprise marketing activities, construct an emoticon dictionary and related corpus. Then, the bullet text and bullet attributes are jointly input into the ALBERT model to extract the feature representation of the bullet text, and fused with the pre trained emoji features of GloVe. Next, using self attention mechanism to capture the relationship between emoticons, bullet text, and bullet attributes, the captured word features are input into BiGRU to capture information in both forward and backward directions, strengthen semantic dependencies, and extract emotional features. Finally, use Softmax logistic regression to classify emotional tendencies and construct an emotional monitoring graph. The performance verification of the model using 163 253 bullet screen data from a certain internet marketing platform shows that the accuracy, precision, and recall rates of the model are 88.8%, 88.7%, and 88.9%, respectively. Compared with other models, the model has improved to a certain extent and can provide support for intelligent and precise marketing of user sentiment monitoring in marketing activities for enterprises.  
      关键词:multi-features;ALBERT model;GloVe model;self-attention mechanism;BIGRU model;sentiment monitoring   
      43
      |
      5
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221663 false
      发布时间:2024-02-22
    • LI Xin,DONG Luyu,SONG Liuguang,SUN Yuqi
      Vol. 23, Issue 2, Pages: 25-31(2024) DOI: 10.11907/rjdk.231112
      摘要:Point cloud registration plays an important role in the development of machine vision, artificial intelligence and other fields. A point cloud registration network (OCADGCNN) based on offset cross-attention is presented to overcome the low accuracy and poor robustness of traditional point cloud registration algorithms and existing deep learning point cloud registration algorithms. The offset attention module is inserted into the Dynamic Graph Convolution Neural Network (DGCNN) to extract global eigenvectors, which can make full use of the local structure and spatial semantics information of point clouds and reduce the loss of information. Include residual connections in feature extraction to improve network performance. The interactive attention module is used to exchange information between global features, enhance related information, and suppress the interference of non-overlapping area information.The experimental results show that the registration effect of OCADGCNN is better than ICP, PointNetLK, PCRNet, OMNet and DOPNet in both noise-free and low-noise ModleNet40 data sets, and the registration accuracy is high. In the experiments of unknown categories, the model has high generalization ability and good versatility, and can better handle low overlap point clouds when the integrity of point clouds is reduced.  
      关键词:point cloud;registration;deep learning;attention mechanism;dynamic graph convolution;feature interaction   
      41
      |
      7
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221897 false
      发布时间:2024-02-22
    • JIANG Bowen,FENG Zijian,HUANG Weihong
      Vol. 23, Issue 2, Pages: 32-39(2024) DOI: 10.11907/rjdk.231054
      摘要:Accurately identifying Transcription factor binding sites in DNA sequences is of great significance for gene expression analysis and drug design. Various prediction methods based on deep learning have been applied to transcription factor binding site tasks, but there is still room for improvement in prediction performance. To this end, a new deep learning method ResNest-TFBS is proposed for predicting transcription factor binding sites on 690 ChIP seq datasets. This method first extracts the spatial structural characteristics of DNA by introducing molecular dynamics features and electrostatic potential energy features based on sequence One-hot encoding; Then, the ResNest model is trained using the split attention mechanism and residual structure to apply the channel attention mechanism to different channel branches, in order to capture the interaction and multi-channel representation of features learned on the global dataset; Finally, the above prior knowledge was transferred to 690 ChIP seq datasets and extensively tested. The experimental results show that ResNest-TFBS has excellent performance, with an average AUC of 0.929. In addition, the SHAP tool was used to verify the contribution of different features in this task, confirming that the introduced features provide more valuable biological clues for predicting transcription factor binding sites.  
      关键词:DNA;transcription factor binding sites;deep learning;transfer learning;split attention mechanism   
      41
      |
      5
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 44816345 false
      发布时间:2024-02-22
    • XIE Jiacheng,JIANG Jianmin,CHEN Huahao,TANG Guofu
      Vol. 23, Issue 2, Pages: 40-47(2024) DOI: 10.11907/rjdk.231964
      摘要:In the process of object-oriented development, UML class diagrams, as the main way to express the static structure of a system, contain a large amount of structural information, making it difficult to ensure consistency between various hierarchical class diagrams in the software development stage. To ensure the consistency of UML class diagrams, formal methods are usually combined with UML class diagrams to analyze and verify the refinement process of class diagrams. However, there is often a problem of information loss or increase in UML class diagram models during the formalization process. Introducing a formal model with a unified structure will not lose or add information. Therefore, first describe the UML class diagram as a unified structure and provide the definition of refinement functions; Then provide an algorithm to verify the effectiveness of the refined function; Finally, perform instance verification on the developed prototype tool. The verification experiment results of the refinement process indicate that this method can help designers timely detect and handle inconsistencies in class diagrams.  
      关键词:UML;class diagram;consistency;refinement;formalization   
      20
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222020 false
      发布时间:2024-02-22
    • HU Jiawei,GUO Liang,ZHOU Yu
      Vol. 23, Issue 2, Pages: 48-54(2024) DOI: 10.11907/rjdk.231217
      摘要:The electric vehicle motor drive system mostly adopts the traditional PI control. Because of the internal saturation limit of the system, the integral saturation phenomenon occurs, and the system is prone to generate the problem of over-harmonic oscillation. Therefore, this paper proposes a parameter improved anti-saturation sliding mode control strategy. First of all, the new strategy uses a smooth non-singular terminal sliding mode controller based on power-exponential mixed reaching law to replace the traditional sliding mode control to reduce the steady-state error and speed up the response speed. Then, a parameter adaptation anti-saturation integrator is designed to calculate the error before and after amplitude limiting and compensate it into the speed loop, so as to reduce the influence of the saturation effect of the PMSM physically limited system on the system overshoot. In addition, Lyapunov theory is used to prove the parameter adaptation anti-saturation sliding mode controller to ensure the stability of the system. The simulation and experimental results show that this strategy can effectively improve the overshoot of the system and weaken the chattering problem of the sliding mode control in the motor speed control.  
      关键词:electric vehicle;permanent magnet synchronous motor;nonsingular terminal sliding mode control;anti-saturation integrator;Lyapunov;DSP   
      21
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 48404505 false
      发布时间:2024-02-22
    • ZHANG Xinyu,ZHANG Yingchao,XUE Weilian
      Vol. 23, Issue 2, Pages: 55-65(2024) DOI: 10.11907/rjdk.232063
      摘要:Aiming at solving the problem of the RFID reader collision problems in the defined area,the G_NSGA-II algorithm is proposed on the basis of the NSGA-II algorithm.A mathematical optimization model is established with the RFID network planning problem, this paper adds a global archiving to perserve elite individuals, replaces the original tournament selection with the tournament selection method with elite preservation, and replaces the polynomial variants with Gaussian variants,to solve the RFID network planning problem.Simulation results show that the G_NSGA-II algorithm produces a better quality solution set, and compared with the three algorithms NSGA-II, AW_GA and MOEA/D, it has an obvious advantage in the distribution and convergence of the solution set. It is effective and feasible in solving RFID problems.  
      关键词:RFID;network planning;NSGA-II algorithm;G_NSGA-II algorithm;reader collision   
      19
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222226 false
      发布时间:2024-02-22
    • CAI Xiaonan,ZHENG Zhongtuan
      Vol. 23, Issue 2, Pages: 66-73(2024) DOI: 10.11907/rjdk.222201
      摘要:Identifying the important nodes in the network has important practical value for studying the topology and functional characteristics of the network. In order to better mine the important nodes in complex networks, considering the impact of the changes of node core location and network topology on the evaluation of node importance, two improved evaluation methods of important nodes in complex networks are proposed based on the node shrinkage method, and simulation experiments and comparative analysis are carried out. On the hand, combined with the characteristic that k-shell value can evaluate the coarse-grained location of nodes, the ratio of k-shell value of nodes to the sum of all k-shell values in the network is taken as the coefficient of node importance (IMC) obtained by node shrinkage method, and an improved algorithm based on the new node importance K-IMC is proposed; On the other hand, the change of network topology is described by network structure entropy. Combined with the change of network standard structure entropy before and after shrinkage, an improved algorithm based on another new node importance E-IMC is proposed. On this basis, simulation experiments are carried out on these two improved important node evaluation algorithms, and the performance of the algorithm is evaluated and analyzed by using SIR model and robustness test. The experimental results show that K-IMC algorithm and E-IMC algorithm show better accuracy in sorting important nodes compared with the original node shrinking method. In terms of accuracy, E-IMC algorithm is higher than K-IMC algorithm, and in terms of operational efficiency, K-IMC algorithm is better than E-IMC algorithm.  
      关键词:complex network;node contraction method;k-shell;standard structure entropy   
      19
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221895 false
      发布时间:2024-02-22
    • TANG Cunhua,TANG Kezong
      Vol. 23, Issue 2, Pages: 74-83(2024) DOI: 10.11907/rjdk.231752
      摘要:To solve the traveling salesman problem, a power law transformation based optimization ant colony algorithm is proposed based on the optimization of ant colony system in ant colony algorithm. Firstly, using power-law transformation to optimize ant colony algorithm to improve the local updating formula of pheromones; Then, in the state transition, a power-law transformation is used to determine the number of times the population has traversed each path, and the impact of local pheromone updates is analyzed through normalization to accelerate the convergence speed of the model; Finally, random addition of Levy flight to disrupt global pheromones prevents the model from falling into local optima too early. Through a large number of instances provided by the TSPLAB database, it has been verified that the power law transformation optimized ant colony algorithm can effectively avoid the model from falling into local optima too early while maintaining a fast convergence speed.  
      关键词:travel quotient problem;ant colony optimization;power-law transformation;Lévy flight   
      26
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222070 false
      发布时间:2024-02-22
    • NING Fanghua,HUANG Bingqi,ZHOU Xiaomin
      Vol. 23, Issue 2, Pages: 84-91(2024) DOI: 10.11907/rjdk.222473
      摘要:To address the complex production planning and scheduling characteristics of the mixed flow production model with multiple varieties and small batches, this paper presents an equal batching strategy to solve the batching problem, so as to realize the simultaneous processing of workpieces in different processes and reduce the machine waiting time. With the maximum completion time as the optimization objective, this paper establishes a mathematical model of the batch scheduling problem in the mixed flow shop; this paper designs an improved genetic algorithm to solve the model, using a combination of the NEH heuristic algorithm and random generation. The improved genetic algorithm is designed to generate high-quality initial solutions using a combination of the NEH heuristic algorithm and random generation, a binary tournament for selection operations, a binary crossover for crossover operations, insertion variation for generating new individuals, a greedy insertion of the domain search algorithm for local search, and a "sub-batch first + idle first" strategy for decoding. The application results of the engine connecting rod production case show that the hybrid flow shop batch scheduling problem model and the improved genetic algorithm are correct and effective.  
      关键词:hybrid flow shop;lot-streaming;genetic algorithm;batch strategy   
      17
      |
      8
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221641 false
      发布时间:2024-02-22
    • SHI Huaiming,ZENG Haoyang,LIANG Guoquan,ZHANG Guofei,XIE Yan
      Vol. 23, Issue 2, Pages: 92-98(2024) DOI: 10.11907/rjdk.222459
      摘要:Under the background of the Internet and big data, The public opinion of drug safety has been paid more and more attention by the government supervision departments because of its wide influence and strong abruptness. However, the data quality of many drug safety public opinion data has not been fully guaranteed, it is difficult for business departments to directly use these public opinion data. In order to better utilize the value of public opinion data, based on the concept of data middle platform, this paper studies solutions for data analysis and processing in the field of drug safety public opinion, and designs a data middle platform with functions of data extraction and transformation, data storage, automatic statistical analysis and visualization of data. It improves data quality and makes it easier for staff to conduct drug safety public opinion analysis and decision-making more effectively.  
      关键词:drug safety;public opinion data;data middle platform;data warehouse;data visualization   
      15
      |
      4
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49212313 false
      发布时间:2024-02-22
    • LOU Zeyao
      Vol. 23, Issue 2, Pages: 99-105(2024) DOI: 10.11907/rjdk.231202
      摘要:Carbon emission forecasting can play an effective role in promoting carbon emission planning and policy promulgation in our country. The wavelet threshold denoising model can remove the noise in carbon emission data, and obtain the effective growth trend of the data. On this basis, the Projection Pursuit Autoregressive (PPAR) model is established by using the time series characteristics of carbon emission data, so as to predict and analyze carbon emission. Compared with the PPAR model without denoising, BP, LSSVM, SVR and LSTM models, the WTD-PPAR model has higher prediction accuracy and more accurate prediction results. The results show that China's carbon emission will reach the peak in 2029, about 1 081.89mt, which can achieve the goal of carbon peak.  
      关键词:carbon emissions;wavelet threshold denoising;projection pursuit;parasitic predator algorithm;autoregression   
      17
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222010 false
      发布时间:2024-02-22
    • XU Jingcheng,CHEN Xuebin,DONG Yanling
      Vol. 23, Issue 2, Pages: 106-112(2024) DOI: 10.11907/rjdk.231070
      摘要:Distributed denial of service (DDoS) attacks are one of the main threats to network security. In recent years, the number of mixed attacks based on various different DDoS attack methods has significantly increased. How to simultaneously detect multiple types of DDoS attacks while ensuring accuracy has become an urgent problem to be solved. To this end, a deep forest based multi type DDoS attack detection method is proposed. This method first uses a feature selection algorithm based on average impure to sort and filter features on multiple types of abnormal traffic datasets; Then, multi granularity scanning is used to extract features from the DDoS training set, and a cascaded forest hierarchical training model is used to generate a deep forest model that can be used for DDoS malicious traffic detection and classification. The experimental results show that compared with six mainstream tree class ensemble learning models, the classifier trained based on the improved deep forest DDoS attack detection method has the lowest accuracy improvement of 0.8% and the lowest recall improvement of 0.9%; Compared with before the improvement, the accuracy of the improved model increased by 1.3%, the weighted recall increased by 1.3%, and the training time decreased by 29.7%. The overall performance of the model has significantly improved.  
      关键词:multi-type attack detection;distributed denial of service attack;deep forest;multi-granularity scanning;cascade forest;average impurity   
      38
      |
      3
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 44816421 false
      发布时间:2024-02-22
    • FANG Wangsheng,CHEN Xiaodong
      Vol. 23, Issue 2, Pages: 113-119(2024) DOI: 10.11907/rjdk.231266
      摘要:To ensure the integrity and security of medical images, a block based semi fragile watermarking medical image algorithm based on singular value decomposition is proposed, which has two functions: locating image tampered areas and image self recovery. This algorithm divides the image by 4 × Divide the watermark into blocks of 4 sizes and divide the embedding position into two types: block authentication bit and self recovery reset. The block authentication bit is used to authenticate each block, ensuring that it can be recognized when tampered with; Self recovery reset is used to achieve self recovery of images after being tampered with. Generate block authentication bits through singular value decomposition for block partitioning, while dividing the block into 2 × Calculate the average value of 5MSB for blocks of 2 to generate self recovery reset. The simulation experiment results show that the PSNR index of the proposed algorithm is above 50 dB, and for most tampering attacks, the Rfa and Rfd are close to 0, and the PSNR index of the self recovered image is above 30 dB. The proposed algorithm has good transparency, can accurately detect most tampering attacks, and has good self recovering image quality.  
      关键词:image security;fragile watermarking;tamper detection;self recovery;medical image   
      16
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221995 false
      发布时间:2024-02-22
    • ZHANG Yao,WANG Xiyin
      Vol. 23, Issue 2, Pages: 120-128(2024) DOI: 10.11907/rjdk.231059
      摘要:A color image encryption algorithm based on hyperchaotic system scrambling, diffusion, and DNA encoding is proposed to address the issues of weak key sensitivity and strong correlation between adjacent pixels in image encryption schemes. Firstly, the original image is divided into 16 sub blocks by performing zero padding operations on the three two-dimensional grayscale image matrices obtained by layering the R, G, and B channels; Then, each module is diffused to obtain a random matrix, and the DNA encoding and operation rules are determined using the four chaotic sequence values generated by the multi chaotic system. DNA operations are performed on the zero filled matrix and the random matrix, respectively. To achieve better diffusion effect, after DNA operation and decoding operation on the two matrix sub blocks, the two chaotic sequences iterated by Logistic chaotic mapping are scrambled in row and column positions, and the image sub blocks are recombined to obtain a color encrypted image. Simulation shows that the improved encryption algorithm achieves a key space of 10128 orders of magnitude, with an average correlation coefficient of -0.004 3 in the horizontal, vertical, and diagonal directions of the image. The calculated UACI value is 33.368 7%, and the NPCR value is 99.695 6%. It has a stronger ability to reduce the correlation between adjacent pixel values in the image and various attacks.  
      关键词:image encryption;hyper-chaotic system;Logistic chaotic map;DNA encoding;DNA operation   
      13
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222199 false
      发布时间:2024-02-22
    • YAO Binrong,ZHANG Na
      Vol. 23, Issue 2, Pages: 129-134(2024) DOI: 10.11907/rjdk.231129
      摘要:Aiming at the problems that the current static detection methods of Android malware have a single type in feature selection, a large number of the same types and low efficiency of the detection model, this paper proposes a static detection method of Android malware based on combined features. The combined feature set consists of three aspects: permission, component and predictability. First, the features of different aspects are selected and retained as the final feature set by means of experiment and reasoning. Secondly, the classification rules of decision tree nodes are optimized according to the information gain of feature attributes, and the detection model is constructed. The experimental results show that the detection accuracy and efficiency of the improved decision tree algorithm model are greatly improved, and the detection results are better than those of the popular random forest algorithm, support vector machine algorithm and naive Bayes algorithm under the same experimental environment.  
      关键词:malware detection;combinatorial features;static analysis;feature set;decision tree algorithm   
      16
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 44815089 false
      发布时间:2024-02-22
    • HU Xinrong,WANG Shenghui,CAI Hao,LUO Ruiqi
      Vol. 23, Issue 2, Pages: 135-145(2024) DOI: 10.11907/rjdk.222360
      摘要:Aiming at the inconsistency problem between the generated image and the semantic layouts in the semantic image synthesis, as well as the unsatisfactory image quality and the irrationality of the combination and collocation of local image, a semantic image synthesis method combining the local-global network model and the instance-adaptive normalization method is proposed. In the global network, the instance-adaptive normalization model is used to stochastically sample the congeneric instance in the network to independently modulate the parameters of each instance, so as to improve the instance matching degree at the global level; build a local generation network, and construct sub generators for each instance according to instance labels, enabling it to capture finer details; the local and global network is combined to fuse local instance image features into the generated image, enabling it to achieve clear instance-level presentation on the global scale. The optimized model has been tested on COCO-stuff, ADE20K, and Cityscapes. The experimental results show that the overall visual effect is more exquisite and realistic, with an average increase of 2.7 MIoU and an average decrease of 5.1 FID.  
      关键词:image generation;image translation;GAN;semantic masks;local generator;instance adaptive   
      17
      |
      6
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221643 false
      发布时间:2024-02-22
    • GAO Deao,CHEN Xiaorong,ZHANG Chiyi,ZU Heyang,GU Xurong,DONG Hongjie,WANG Yuanji
      Vol. 23, Issue 2, Pages: 146-152(2024) DOI: 10.11907/rjdk.231076
      摘要:Wafer chip detection plays a crucial role in the wafer processing and production process. To address the limitations of long time and low accuracy in wafer chip detection in industrial production processes, an improved multi-objective template matching algorithm based on machine vision combined with non maximum suppression algorithm is proposed. This algorithm utilizes the nearest neighbor bounding rectangle algorithm to obtain the rectangular contour that best fits the chip, accurately obtaining the template of the rectangular chip; For chip surface contamination that affects template matching, a grayscale compensation method combined with morphological improvement is adopted to reduce the impact of grayscale values in the contaminated area on the matching results. The experimental results show that the recognition rate of the proposed multi-objective template matching algorithm is over 95%, and the time consumption does not exceed 0.5 seconds; The nearest neighbor bounding rectangle algorithm is more accurate than the traditional minimum bounding rectangle algorithm, providing a feasible solution for industrial wafer chip detection.  
      关键词:wafer chip detection;machine vision;non-maximum suppression;multi-object template matching;near-neighbor external rectangle;grayscale compensation   
      22
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 44816379 false
      发布时间:2024-02-22
    • LAI Yibo,YU Qingcang,FANG Jiaji,JIANG Lurong,WU Yao,HUANG Zheng
      Vol. 23, Issue 2, Pages: 153-161(2024) DOI: 10.11907/rjdk.231141
      摘要:In agricultural grafting cultivation, it is usually necessary to ensure that the scion leaves and rootstock leaves are cross-shaped after grafting. In order to enable the automatic grafting machine to accurately segment the scion leaves in real time and find out the characteristic parameters of the cotyledons, a lightweight segmentation method based on improved UNet is proposed, Using the MobileNetV2 backbone as the feature extraction backbone, the Ghost Module is used to implement double convolution operations in the enhanced feature extraction layer, which improves network accuracy while reducing network parameters and calculations. The experimental results show that compared with the original model, the Mobile-UNet model has increased by 5.69%, 1.32%, 4.73% and 3.12% in indicators such as MIoU, Precision, Recall and Dice coefficients, and the calculation amount and parameter amount of the model have decreased by 27.4% and 35.3%. In addition, compared with SegNet and DeepLabV3+ classic segmentation models, this model has higher segmentation accuracy and fewer parameters. It is deployed in the automatic grafting machine system to realize the segmentation of scion cotyledons on the clamping mechanism.  
      关键词:grafting machine;scion leaf;improved UNet;Ghost module;semantic segmentation   
      24
      |
      4
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221785 false
      发布时间:2024-02-22
    • LI Zheng,QIAO Baojun,YANG Wei,YUAN Caihong,LIU Cheng
      Vol. 23, Issue 2, Pages: 162-166(2024) DOI: 10.11907/rjdk.231084
      摘要:Under the background of engineering education accreditation, in order to carry out “student-centered, outcome-oriented” data structure teaching, with the help of modern information technology, a multi-dimensional integrated teaching mode of online and offline, in-class and out-of-class, as well as theory and practice has been constructed. Guided by OBE, this teaching mode carries out blended teaching, teaching methods and assessment design. Based on flipped classroom, it effectively integrates learning activities of different dimensions, stimulates students’ initiative in independent learning, and expands their knowledge system. At present, this teaching mode has been applied to the computer science and technology major of Henan University for three semesters. From the feedback of students and experts as well as the achievement of course objectives, the teaching effect is obvious. The multi-dimensional integration teaching mode effectively improves students’ autonomous learning ability, ability to analyze and solve practical problems and expand innovation capability.  
      关键词:OBE;data structure;blended teaching;multi-dimensional assessment   
      21
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221803 false
      发布时间:2024-02-22
    • DAI Jianfeng
      Vol. 23, Issue 2, Pages: 167-171(2024) DOI: 10.11907/rjdk.232255
      摘要:With the continuous development of society, higher requirements have been put forward for electrical engineering talents in the context of the new power system. Starting from the current problems in the cultivation of electrical engineering talents, this paper delves into the challenges of a single teaching process, a focus on theory over practice, and formal assessment. To address these issues, a teaching model of industry oriented academia research is proposed, which emphasizes curriculum design and updates, and implements key measures such as a hierarchical integrated teaching system, industry academia research teaching model, and improved curriculum assessment system. Practice has proven that curriculum innovation can better cultivate composite talents in electrical engineering with innovative abilities and practical skills, enabling them to better adapt to the development trend of new power systems.  
      关键词:new power system;electrical engineering;industry-university-research teaching;practical ability   
      20
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222232 false
      发布时间:2024-02-22
    • LIU Jiang,ZHANG Xiaoqing,WU Xiao,XIAO Zunjie,HU Lingxi
      Vol. 23, Issue 2, Pages: 172-176(2024) DOI: 10.11907/rjdk.231681
      摘要:Recently, artificial intelligence and natural language understanding technology have gradually affected the traditional computer course teaching mode, where ChatGPT is one representative example. To cope with the challenges and opportunities brought by ChatGPT and related artificial intelligence technology, taking the course "multimedia information processing (MIP)" of Southern University of Science and Technology as an example, this paper proposes the flexible "ChatGPT+"-driven innovative method of course teaching, elaborating its practice and exploration. The feedback from students through the questionnaire indicates that this course innovation teaching design improves the teaching effect and learning interest of students. This paper hoped that it can provide a reference for the traditional computer course teaching mode.  
      关键词:multimedia information processing;teaching innovation;ChatGPT;active learning;collaborative learning   
      56
      |
      6
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222201 false
      发布时间:2024-02-22
    • LI Xiangli,LI Lei
      Vol. 23, Issue 2, Pages: 177-181(2024) DOI: 10.11907/rjdk.231106
      摘要:The rapid development of computer network discipline can easily lead to the teaching content of advanced computer network courses lagging behind the development of network technology. In response to the problems in curriculum construction and the emphasis on theory over practice in the teaching process, based on the actual situation of the course, it is proposed to establish a curriculum system based on network theory, with cutting-edge technology and applications as the content, and cultivating scientific research ability as the core. Analysis, exploration, and practice were conducted from the aspects of teaching content and methods, practical teaching methods, paper discussions, teaching resource sharing, and course assessment mechanisms. Teaching practice has shown that the constructed advanced computer network course system not only cultivates students' professional literacy and scientific thinking methods, but also enhances their practical and innovative abilities, laying a foundation for their later scientific research work.  
      关键词:advanced computer network;curriculum construction;practical teaching;reform in education;course assessment   
      11
      |
      1
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49211766 false
      发布时间:2024-02-22
    • ZHU Shiwei,YE Baolin,WU Weimin
      Vol. 23, Issue 2, Pages: 182-193(2024) DOI: 10.11907/rjdk.222426
      摘要:In recent years, thanks to the upgrading of traffic detection equipment and urban data storage infrastructure as well as the rapid development of deep learning technology, the application of deep learning technology to solve the problem of short-term traffic flow prediction has become a research hotspot in the field of intelligent transportation. Different from the traditional short-time traffic flow prediction methods, the short-time traffic flow prediction method based on deep learning can make full use of massive traffic data, dig deeply the hidden features and associations between traffic nodes, and effectively improve the accuracy of short-term traffic flow prediction. Firstly, this paper briefly reviews the development history of short-term traffic flow prediction methods, and focuses on analyzing and discussing the latest technical progress and theoretical research results of short-term traffic flow prediction methods based on deep learning model. Then, the open traffic flow data sets, which are widely used to verify the effectiveness of the algorithm and make comparative analysis, are combed and summarized. In addition, the specific process and detailed steps of applying the short-time traffic flow prediction algorithm based on the deep learning model to solve the actual traffic flow prediction problem are described. The short-term traffic flow prediction algorithm based on the deep learning model LSTM and GRU is simulated with the open test data set PEMS04. Simulation results verify the effectiveness of the algorithm and its advantages compared with traditional methods. Finally, the challenges and future development directions in the field of short-term traffic flow forecasting have been summarized and prospected.  
      关键词:short-term traffic flow prediction;deep learning;time series;traffic dataset;convolutional neural networks   
      27
      |
      6
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221675 false
      发布时间:2024-02-22
    • HAN Meng,JIAO Jianlin,LIU Sichun,WANG Song,LIU Kai,JI Yuejin
      Vol. 23, Issue 2, Pages: 194-200(2024) DOI: 10.11907/rjdk.221982
      摘要:The multipoint control unit resource (MCU) scheduling method is widely used in practical application scenarios of video communication system and has an important impact on the overall performance and operation development of video communication. In this paper, the existing MCU resource scheduling methods are sorted out. First, the MCU resource scheduling process in video communication is introduced, and the objectives and characteristics of MCU resource scheduling are summarized; Then, the existing MCU scheduling algorithms are classified and discussed, and the advantages and disadvantages of various scheduling methods are compared; Finally, the results and problems of the existing scheduling research are discussed, and the future research is addressed. The MCU resource scheduling of video communication can be more efficient by realizing the cloud service of MCU equipment, which is conducive to the performance improvement and long-term stable development of video communication.  
      关键词:video conference system;MCU resource scheduling strategy;algorithm optimization   
      15
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49222230 false
      发布时间:2024-02-22
    • NI Jia,ZHANG Peiming,HU Tong,SHEN Jialin,WANG Ying,GUO Shijun
      Vol. 23, Issue 2, Pages: 201-207(2024) DOI: 10.11907/rjdk.231126
      摘要:This paper reviews smartphone-based ophthalmic diagnostic technologies and analyzes the technical features of slit lamp microscopes, fundus imaging devices and corneal topographers that use smartphone optical imaging technology for ocular structural examination. From the research,the application of smartphones in ophthalmology has the advantages of good portability, ease of use and low cost, which can increase the opportunities of patients, especially patients in areas with scarce medical resources, to have access to ophthalmic disease diagnosis, and can save professional medical resources. In the future, smartphone-based ophthalmic diagnostic devices will have a wide range of applications in hospitals, clinics and community health centers for ophthalmic disease screening, and can also be used in medical schools for clinical teaching.  
      关键词:ophthalmic diagnosis;smartphone;slit lamp microscope;corneal topographer;ophthalmoscope   
      16
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221696 false
      发布时间:2024-02-22
    • WANG lei,LU Shanshan,ZHANG Songlan
      Vol. 23, Issue 2, Pages: 208-214(2024) DOI: 10.11907/rjdk.222182
      摘要:Industrial software is called the brain and nerve of industrial manufacturing. Due to the dominance of European and American software in the market, industrial software has become a weak area in China's industrial sector. Niche strategy is based on the current situation of domestic industrial software, which helps it find its own development orientation. Based on the niche strategy theory, try to reduce the competitive pressure of domestic industrial software from multiple angles. And through the case of Chuangyuan's digital transformation of Beilun mold industry, the applicability of China's industrial software ecological niche development strategy is analyzed. Provide reference for domestic industrial software enterprises, so that they can find the right development space and promote the sustainable and healthy development of domestic industrial software.  
      关键词:niche strategy;domestic industrial software;niche overlap;niche breadth;Industrial Internet   
      13
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221997 false
      发布时间:2024-02-22
    • WANG Shuo,CHENG Yunzhang
      Vol. 23, Issue 2, Pages: 215-220(2024) DOI: 10.11907/rjdk.231145
      摘要:With the development of artificial intelligence technology, the recognition effect of deep learning in gesture recognition has been significantly improved. Surface electromyography (SMG) signal is an electrophysiological signal generated during muscle activity in the human body. Due to its non-invasive and easy to collect nature, it has been used as a signal source for rehabilitation aids and prosthetic control. When applying surface electromyography signals, pre-processing such as amplification and filtering is required; Then, feature extraction is carried out to obtain effective information of surface electromyography signals in the time domain, frequency domain, and timely frequency domain; Finally, by inputting this information into the machine learning model, the relevant muscle movements of the human body can be analyzed, and then the movements of the relevant instruments can be controlled. To this end, a review is mainly conducted on the feature extraction and machine learning classification models, elaborating on the current research progress and future development direction of gesture recognition based on surface electromyography signals.  
      关键词:surface electromyography signal;feature extraction;machine learning;deep learning;gesture recognition   
      18
      |
      0
      |
      0
      <HTML>
      <L-PDF><WORD><Meta-XML>
      <引用本文> <批量引用> 49221673 false
      发布时间:2024-02-22
    0