Research On Safety Detection Technology Of Power Live Working Based On Deep Learning

The work regulations of electric power production safety regard “safety first and prevention first” as the main policy of power production safety.Therefore,it is necessary to strengthen the on-site monitoring of electric power operations and timely detect and eliminate potential safety hazards to ensure stable production and personnel safety.With the application of convolutional neural networks in the field of computer vision,vision-based target detection and tracking technology has made great progress.The application of target detection and tracking technology in the field of power production safety to realize intelligent safety monitoring has broad research prospects and application potential.Therefore,this article tries to improve and innovate related technology on the basis of the existing target detection and tracking technology,carrying out a series of research work with the aim of practical application.The achievements are as follows:(1)With the changeable scale of target in the power operation scenario,the accuracy and real-time performance of the detection method must meet the requirements of practical applications.After researching,analyzing and comparing two types of models of R-CNN series and YOLO series,the following conclusions are drawn: YOLOv3 algorithm has the characteristics of multi-scale prediction,the most balanced speed and accuracy,so it is the most suitable for target detection in power operation scenarios.(2)This article proposed an anti-occlusion tracking method based on multi-feature fusion and adaptive learning rate updatint.It adds the strategy of fusion feature,multi-scale detection,and adaptive learning rate based on occlusion detection,which makes up for the shortcomings of the original KCF algorithm,such as single feature,inability of scale adaptation and blind updating.The test results on the OTB-2013 tracking data set show that the improved algorithm has excellent accuracy and robustness.(3)This article proposed a multi-target tracking method based on priority and motion estimation for target association.Considering the target occlusion problems in multi-target tracking,Kalman tracking is added in the tracking part as a supplementary output when the target is occluded.Aiming at the problem of target association,this paper proposed an association method based on priority and motion estimation to match the detected target with the tracking target.Furthermore,the multi-target tracking method proposed in this paper and the classic Sort algorithm are tested and compared on the evaluation data set.The results show that the multi-target tracking method proposed in this paper performs better in performance indicators such as tracking accuracy,missed detection,and target ID exchange.(4)The YOLOv3 algorithm and the multi-target tracking method proposed in this paper are used for safety detection in electric power live working places,which mainly includes the tracking and trajectory recording of workers,the detection of the helmet wearing state and the detection of mis-entry into the dangerous area.The test results show that the accuracy of the method in this paper can reach 96.6% and 95.6% in the detection of helmet wearing status and the mis-entry in dangerous areas in the power operation scenario respectively.