Computer and Modernization

Group Activity Recognition Algorithm Based on Interaction Relationship Grouping Modeling Fusion

WANG Chuan-xu, LIU Ran

2022, 0(01): 1-9.

Asbtract ( 448 )

PDF (3063KB) ( 429 )

References | Related Articles | Metrics

The modeling of interaction relationship between group members is the core technology of group activity recognition. High complexity and information redundancy in relational reasoning are tough problems in complex scenarios when modeling its group interactions. In order to solve these problems, we propose a model of grouping interactive relation. Firstly, CNN and RoIAlign are used to extract the scene information and personal information as initial features in each frame, and the whole group is divided into two subgroups by the personal spatial coordinates (For example, in the Volleyball data set, the X coordinates of participants’bounding boxes are used to rank, then, everyone set is set up an ordinal ID and 12 people are divided into two group from left to right). Secondly, the two local groups and the global scene groups are divided, the Graph Convolutional Network （GCN） is used to deduce their interaction relationship respectively, and the key persons in each group are determined. Then, we can regard global relationship features as the real value, and merge the characteristics of local relation of two groups as predicted value. In order to match the key figures of two groups with key figures from the whole group successfully, the cross-entropy loss function is built between the two and feedback to optimize the upper-level group GCN interaction relationship network. Next, with the information of key figures in the global interaction relationship as a guide, the key figures in the two subgroups are matched respectively. After successful matching, the matched key figures in the two subgroups are taken as the target nodes to establish a relationship graph between these two subgroups, and then it is deduced by GCN. Finally, the initial features are fused with intergroup and global interaction characteristics respectively to obtain two group behavior branches, and the final recognition result is obtained through decision fusion. The experiment shows that the accuracy is 93.1% on Volleyball data set and the accuracy is 48.1% on NBA data set.

Relationship Extraction Method Based on BiLSTM and ResCNN

XU Xiao-liang, ZHAO Ying

2022, 0(01): 10-16.

Asbtract ( 369 )

PDF (1310KB) ( 335 )

References | Related Articles | Metrics

Most of relationship extraction methods cannot obtain long-distance dependent information from long sentences, and the performance of relationship extraction is degraded due to the data noise. This paper proposes a new relationship extraction model based on BiLSTM and ResCNN to solve these problems. The model uses BiLSTM to obtain the context information vector of words. The features of the middle or low layer in the convolutional neural network are transferred to the high layer through residual network, which effectively solves the problem of vanishing gradient. At the same time, embedding the squeeze-and-excitation block into the residual network can greatly reduce the data noise and strengthen the feature transfer. The piecewise max pooling method is used to capture the structural information of the entity pair. This paper designs verification experiments on NYT-Freebase dataset. Experimental results show that this model can fully learn features and significantly improve the effect of relationship extraction.

Fast Memory Synchronization Technology for Container Thermal Migration

YOU Qiang-zhi, HU Huai-xiang, CHEN Xiang-yu

2022, 0(01): 17-22.

Asbtract ( 322 )

PDF (944KB) ( 290 )

References | Related Articles | Metrics

Container hot migration is the basis of cloud platform load balancing technology, and also an important guarantee for cluster fault management and underlying system maintenance. At present, the implementation of container hot migration is mainly based on checkpoint/restore mechanism， that is to do checkpoint operation on the running container, then stop the container, transfer the image file to the destination host, and then recover. The migration time includes checkpoint time, transmission time and recovery time. In order to reduce the downtime of container hot migration and reduce the transmission consumption, this paper designs and implements a container hot migration scheme based on pre-copy migration algorithm, and adopts the key technology of fast memory synchronization, which includes three methods: fine-grained dirty memory identification, dirty memory compression and transfer, and merging incremental memory in advance. Experiments show that the proposed scheme and optimization technology can significantly reduce the downtime and transmission overhead.

Risk Prediction Model of Heart Failure Unplanned Readmission Based on ADE-Stacking

WANG Lei, SONG Bo

2022, 0(01): 23-27.

Asbtract ( 188 )

PDF (1039KB) ( 104 )

References | Related Articles | Metrics

With the increasing aging of the population, the incidence of heart failure has increased, and the problem of unplanned readmission of patients with heart failure has led to a decrease in the quality of life of patients and an increase in medical costs. Therefore, it has become an urgent problem to be solved. Aiming at the problem of readmission risk prediction, this paper proposes an unplanned readmission risk prediction model for heart failure patients based on ADE-Stacking. This model is mainly composed of two parts: integrated learning algorithm model construction and parameter optimization. The integrated learning algorithm can be combined with multiple parts. The advantages of a weak classifier make the model have better generalization and accuracy. The parameter optimization part uses the adaptive shrinkage factor F to improve the differential evolution algorithm to improve the parameter optimization performance. The model is trained and tested using the heart failure readmission patient data set. The results show that the proposed model is better than other machine learning algorithms such as random forest, XGBoost, support vector machine and other commonly used risk prediction models.

Hybrid Recommendation Algorithm for BIM Model Based on Model Feature Matching

XIAO Hong-yu, ZENG Wen-qu, WANG Shu-ying

2022, 0(01): 28-32.

Asbtract ( 212 )

PDF (757KB) ( 158 )

References | Related Articles | Metrics

In order to assist the professional designers of subway engineering to quickly obtain the reference model matching the current design requirements from the BIM model case library, a BIM model hybrid recommendation algorithm based on feature matching is proposed. Firstly, the feature data is obtained from BIM model based on the secondary development of Revit. Secondly, the entropy weight grey correlation model is used to calculate the recommendation degree of the model instance by using the basic information such as the model feature parameters. Then, the recommendation degree of the model instance is calculated by using the fusion model of gradient boosting decision tree algorithm (GBDT) and logical regression (LR) algorithm combined with the user interaction data. Finally, the combination proportion of the two recommenders can be adjusted dynamically according to the scale of the training data set. Experiments show that this method not only avoids the problem of cold start, but also has better BIM model recommendation quality with the support of enough user interaction data.

Particle Swarm Optimization Algorithm Based on Multigroup Parallel Cooperation

GUO Cheng, ZHANG Wan-da, WANG Bo, WANG Jia-fu

2022, 0(01): 33-40.

Asbtract ( 367 )

PDF (3326KB) ( 163 )

References | Related Articles | Metrics

Aiming at the problem that high-dimensional complex optimization problems are prone to dimension disaster, which makes the algorithm easily fall into local optimization, a particle swarm optimization algorithm based on multigroup parallel cooperation is proposed, which can comprehensively consider the characteristics of high-dimensional complex optimization problems and dynamically adjust the evolution strategy. Based on the analysis of the characteristics of particles in the process of solving high-dimensional complex problems, the network model of multigroup parallel cooperation of particle swarm optimization algorithm (PSO) which integrates ring topology, fully connected topology and von Neumann topology is established. The model combines the advantages of three kinds of topology particle swarm optimization algorithm in solving high-dimensional complex optimization problems, designs a multigroup particle broadcast feedback dynamic evolution strategy, and designs an evolutionary algorithm to realize the dynamic adaptation of topology in high-dimensional complex optimization environment, so that the algorithm has strong search ability in solving high-dimensional unimodal function and multi-modal function. The simulation results show that the algorithm has good performance in the optimization accuracy and convergence speed of solving high-dimensional complex optimization problems.

Review of Big Data Workflow Orchestration and Management System in Cloud Environment

CAO Yu, LI Xiao-hui, LIU Zhong-lin, JIA He, FEI Zhi-wei

2022, 0(01): 41-53.

Asbtract ( 232 )

PDF (1624KB) ( 126 )

References | Related Articles | Metrics

With the increasing complexity of big data analysis and processing requirements, the expression of the analysis and processing process needs to be transformed into the form of a big data workflow constructed based on tasks and inter-task dependencies in order to achieve its structured, repeatable, controllable, scalable and automated execution. The issue of big data workflow orchestration and management has become an important research topic. The heterogeneity of resources in the cloud computing environment has made this problem more complicated. This paper first divides the research contents on big data workflow orchestration and management in the cloud environment into four aspects, big data workflow composition, workflow fragmentation, task scheduling and execution, and fault tolerance, and on this basis, it reviews and introduces classic and highly concerned researches in recent years each aspect; then, it classifies and sorts out the mainstream technologies in these researches, and analyzes the methods proposed in each research and their characteristics, advantages, and items to be improved. Finally, the perspective is returned to the big data analysis and processing system, and the benefits of various studies to the system are classified and analyzed.

Impact Point Detecting Algorithm Based on Salient Object Detection

ZHOU Xuan, ZHU Su-lei, HE Wei

2022, 0(01): 54-60.

Asbtract ( 240 )

PDF (1990KB) ( 155 )

References | Related Articles | Metrics

Aiming at the problems of hidden danger, low efficiency of manual measurement and poor accuracy of current projectile flame location methods, an improved projectile flame detection algorithm based on salient target detection network BASNet (Boundary-Aware Salient Object Detection) is proposed in this paper. Using the improved BASNet network, combined with attention mechanism module-CBAM (Convolutional Block Attention Module), pyramid pooling module-PPM (Pyramid Pooling Module) and depth separable convolution, to detect the projectile fire and extract the coordinates of impact point on the image. The experimental results show that detection accuracy of F measure reaches 0.914, the mean absolute error-MAE reaches 0.006 and detection speed reaches 3.86 fps,better than other salient object detection network like BASNet, U2Net. The error between the coordinates of impact point image extracted by this method and the real coordinates is 5.92 pixels, which is 4.85 pixels less than BASNet. In conclusion, the improved network retains the effective shallow semantic information, enhances the detection accuracy of the network for the significant objects, improves the efficiency of model reasoning, and is suitable for the detection of small-scale projectile fire smoke in the shooting range, which can meet the actual needs of the application in the shooting range.

Multilevel Thresholding Image Segmentation Using Improved Pathfinder Algorithm

WANG Shu-ping, LI Min, DU Min, LUO Jian-wei

2022, 0(01): 61-69.

Asbtract ( 190 )

PDF (2228KB) ( 135 )

References | Related Articles | Metrics

There are some problems in multilevel threshold image segmentation, such as large amount of computation and long running time. A new multilevel threshold image segmentation method named improved pathfinder algorithm (IPFA) is proposed using Tent map and adaptive t-distribution strategy on the standard of pathfinder algorithm (PFA). This method uses Kapur’s entropy as the objective function to search the best segmentation threshold. In order to verify the effectiveness of the algorithm, the convergence accuracy and speed of IPFA are tested by benchmark functions at first. Then IPFA-Kapur is applied to multilevel threshold image segmentation and compared with standard PFA, moth-flame optimization (MFO), gray wolf optimization (GWO) and particle swarm optimization (PSO). Experimental results show that the proposed algorithm has faster convergence speed and higher segmentation accuracy, and has better segmentation effect than other comparison algorithms, and the peak signal to noise ratio (PSNR) and structural similarity (SSIM) have better performance, which can effectively solve the problem of multilevel threshold image segmentation.

Lightweight Infrared HV Bushing Identification Algorithm Based on Attention Mechanism

GUO Teng-fei, ZHANG Ze-yan, FU Hong-cai, WANG Ji-xuan, NIU Tian-bao

2022, 0(01): 70-76.

Asbtract ( 174 )

PDF (5759KB) ( 105 )

References | Related Articles | Metrics

In order to improve the identification accuracy of transformer HV bushing in infrared images and meet the needs of mobile terminal and other low-end devices for target detection network, this paper proposes an improved lightweight infrared HV bushing identification algorithm, using Tiny YOLOv3 target detection network as the basic detection network. First, through the fusion of the Convolutional Block Attention Module (CBAM) attention mechanism, the channel attention and the spatial attention mechanism are connected in series to increase the receptive field of the target detection network, while reducing network computing tasks and improving network performance. Then, GIoU loss and Focal loss are used to replace the original bounding box loss and confidence loss, thereby improving the recognition rate of the HV bushing in the infrared image and reducing the occurrence of missed and false detections. The experimental results show that compared with the original Tiny YOLOv3 network, the improved network increases mAP to 96.28%, increases F1 to 96.25%, and the weight size is 33.9 MB, less than that of YOLOv3 training network. It is better suitable for low-end equipment and provides favorable conditions for a smart substation online monitoring.

An Improved Instrument Detection Algorithm Based on YOLOv3

HUANG Zi-ping, HUANG Ji-feng, ZHOU Xiao-ping

2022, 0(01): 77-84.

Asbtract ( 257 )

PDF (2642KB) ( 173 )

References | Related Articles | Metrics

Instrument detection is an indispensable part of intelligent instrument testing, its effect directly determines the accuracy of instrument testing. In view of the complex positioning background of the instrument and the requirement of fast detection speed, a target detection algorithm based on improved YOLOv3 is proposed. Based on YOLOv3 algorithm, the last two network blocks in the Darknet are first replaced with DenseNet (Densely Connected Convolutional Networks) so as to enhance the reuse of features by the model. And then the lightweight Darknet-48 is used as feature extraction networks, and the convolution neural network in the DenseNet is modified to the deep separable convolution network, and then the 6 layer convolution before all detection layers (YOLO Detection) is modified to 2 layers so as to reduce the parameters of the model. At the same time, GDIOU bounding box is introduced to regress coordinates loss, and the weight of the loss function is readjusted according to the detection requirements. Experimental results show that compared with the original algorithm, the number of parameters of the improved YOLOv3 algorithm is reduced by 40%, and the accuracy and recall in instrument detection reach 95.83% and 94.98%, respectively, which is increased by 2.21 percentage points and 2.09 percentage points. The average accuracy is increased by 2.42 percentage points and the detection speed is increased by 30.18%.

Mask Wearing Detection Algorithm Based on Improved YOLOv4

JIN Xin, ZENG Si-ke, LIU Yang, WU Chu-han

2022, 0(01): 85-90.

Asbtract ( 295 )

PDF (3871KB) ( 243 )

References | Related Articles | Metrics

In order to solve the problems of low detection speed and large amount of model parameters of YOLOv4 in the target detection task, an improved target detection algorithm of YOLOv4 is proposed. CSPDarknet53 of the YOLOv4 backbone is replaced by Mobilenet to improve the feature extraction network of YOLOv4, and the original standard 3×3 convolution of PANet is replaced by a depth-division convolution to reduce the computational burden, so as to improve the detection speed and reduce the model parameters. The K-means+〖KG-*3〗+ algorithm is then used to perform anchor dimension clustering on a dataset consisting of 8565 images to improve the accuracy of the algorithm. At the same time, a system for recording pedestrian wearing of masks and measuring people’s temperature is built to perform epidemic control tasks in crowded places. The FPS has been improved by 200% and the model parameters have been reduced by 82% compared with the original algorithm, while maintaining the accuracy of the YOLOv4-Mobilenet. The improved model can detect an average of 67 frames per second, which can detect mask wearing in real applications, and the results show that the model is efficient and reliable.

Industrial Safety Helmet Detection Algorithm Based on Depth Cascade Model

YANG Zhen, ZHU Qiang-qiang, PENG Xiao-bao, YIN Zhi-jian, WEN Hai-qiao, HUANG Chun-hua

2022, 0(01): 91-97.

Asbtract ( 199 )

PDF (13967KB) ( 153 )

References | Related Articles | Metrics

In industrial production, safety helmet provides a better safety guarantee for human head. In the field environment, the inspection of whether the construction personnel wear safety helmet mainly depends on manual inspection, so the efficiency is very low. In order to solve the problem of helmet detection and identification in construction site, this paper proposes a helmet detection method based on the deep cascade network model. Firstly, the construction personnel are detected through the You Only Look Once version 4 (YOLOv4) detection network. Then, the attention mechanism residual classification network is used to classify and judge the ROI region of personnel and identify whether they wear a helmet or not. This method is carried out in the experimental environment of Ubuntu18.04 system and Pytorch deep learning framework, and training and testing experiments are carried out in the self-produced helmet data set. The experimental results show that compared with YOLOv4, the safety helmet recognition model based on the deep cascade network has an accuracy increase of 2 percentage points, which effectively improves the safety helmet detection effect of construction personnel.

Multi-UAV Power Inspection Task Planning Technology Based on Deep Reinforcement Learning

MA Rui, OUYANG Quan, WU Zhao-xiang, CONG Yu-hua, WANG Zhi-sheng

2022, 0(01): 98-102.

Asbtract ( 354 )

PDF (1806KB) ( 571 )

References | Related Articles | Metrics

UAVs have been widely used in the inspection tasks of power grid lines and electrical towers due to their advantages of flexibility, low cost and strong maneuverability. Because of the limited range of a single UAV, multiple UAVs are required to cooperate in a wide range grid inspection. However, the traditional planning methods cannot work well because of slow computing speed and unobvious collaborative effect. To remedy these deficits, a new mission planning algorithm is proposed in this work, which is based on multi-agent reinforcement learning algorithm QMIX. On the basis of the framework of intensive training and decentralized execution, this algorithm establishes RNN network for each UAV and gets the joint action value function guideline for training by mixing network. To simulate the collaboration capabilities of multi-agents, a reward function for collaboration task is designed, and it solves the problem of low collaboration efficiency in multi-UAV mission planning. The simulation results demonstrate that the proposed algorithm takes 350.4 seconds less than VDN algorithm.

Mobile Robot Path Planning Based on Fusion of Improved A* and DWA Algorithms

PANG Yong-xu, YUAN De-cheng

2022, 0(01): 103-107.

Asbtract ( 1316 )

PDF (2046KB) ( 1021 )

References | Related Articles | Metrics

Aiming at the path planning requirements of mobile robot to achieve global optimal path in complex environment and dynamic and real-time obstacle avoidance in unknown environment, the traditional A* (A-star) algorithm is improved, and Dynamic Window Approach (DWA) is integrated to achieve dynamic and real-time obstacle avoidance. Firstly, the obstacle proportion in the grid environment is analyzed. The obstacle proportion is introduced into the traditional A* algorithm to optimize the heuristic function h(n), so as to improve the evaluation function f(n) and improve its search efficiency in different environments. Secondly, in view of the intersection between the trajectory and the vertex of obstacles optimized by the traditional A* algorithm in the complex grid environment, the selection method of child nodes is optimized, and the redundant nodes in the path are deleted to improve the smoothness of the path. Finally, Dynamic Window Approach is integrated to realize dynamic and real-time obstacle avoidance of mobile robot in complex environment. The comparative simulation experiments under MATLAB show that the improved algorithm is optimized in the path length, path smoothness and elapsed time, meets the global optimal and can realize dynamic and real-time obstacle avoidance, and has better path planning effect.

Optimization of Container Cloud Resource Allocation Based on Genetic Algorithm

XU Sheng-chao, XIONG Mao-hua

2022, 0(01): 108-112.

Asbtract ( 219 )

PDF (784KB) ( 275 )

References | Related Articles | Metrics

This paper proposes a genetic algorithm approach for resource allocation optimization in container-based cloud environment. Considering resource allocation when VMS are configured on physical hosts and containers are configured on VMS, the objective function is to minimize the overall energy consumption of the container cloud platform data center. The machine should correspond to the container and other constraints, and the genetic algorithm is used to solve the objective function through five steps of chromosome expression, initialization, crossover operation, mutation operation and setting fitness function to obtain the optimal virtual resource allocation result. The experimental results show that the proposed method can realize the reasonable allocation of virtual resources in the container cloud environment and keep the energy consumption of the container cloud platform data center to a minimum and achieve the goal of resource efficient utilization.

Rumor Source Detection Based on Extended Epidemic Model

WU Yang１, WU Guo-wen１, ZHANG Hong１, SHEN Shi-gen２, CAO Qi-ying

2022, 0(01): 113-119.

Asbtract ( 268 )

PDF (1868KB) ( 256 )

References | Related Articles | Metrics

In order to study the issue of rumors detection of better fitting the actual situation, this paper considers the ability of banning and isolating nodes that spread rumors in social networks,proposes a new model called SIOR (Susceptible-Infected-isOlated-Removed), which is based on the classic model called SIR. Then this paper obtains the source estimator through the optimal information propagation process and verifies that the estimated value is similar to the Jordan Infection Center in the network topology based on the SIOR model. Finally, this paper proposes a reverse infection propagation algorithm for the SIOR model, which can identify the Jordan infection center in the network topology,then compares the algorithm with other centrality detection algorithms through simulation experiments to verify the feasibility of the estimator. In addition, the accuracy under SIOR model is improved compared with SIR model.

Certificateless Signcryption Scheme Based on Blockchain

ZHANG Tian-xi, WANG Li-peng, FU Jun-jun, CUI Ci, JIN Meng-lu

2022, 0(01): 120-126.

Asbtract ( 300 )

PDF (969KB) ( 315 )

References | Related Articles | Metrics

Signcryption algorithm can realize encryption and signature functions at the same time in a logical step. Compared with the traditional scheme of signing first and then encrypting, signcryption algorithm has the advantages of low calculation amount and communication cost, and is widely used in electronic payment, Internet of Things, etc. Existing signcryption schemes based on elliptic curve and bilinear pairing generally have the problem of low execution efficiency. Therefore, this paper proposes a certificateless signcryption scheme based on blockchain. The new scheme is implemented based on discrete logarithm puzzle and has the advantages of high execution efficiency. The new scheme also takes advantage of the non-tamperable modification and traceability of the blockchain to achieve non-repudiation. Security analysis shows that the proposed scheme has the characteristics of non-repudiation, confidentiality, and unforgeability. Performance analysis shows that the new scheme is more efficient in execution. Simulation experiments show that the introduction of blockchain has little effect on the overall performance of the system.

Table of Content