Multi-fault Detection And Tracking Of Transmission Lines Based On YOLO-V4

In the process of power network construction in China,it is always difficult to inspect transmission lines manually.In today’s highly developed UAV technology,many areas are trying to apply UAV to artificial patrol.Although the unmanned aerial vehicle can effectively improve the efficiency of power line patrol to some extent,the original pictures are not the result of analysis for defects.Such pictures still need to be analyzed manually to find various problems in the circuit.To solve this problem,a multi-fault detection and tracking system for transmission lines based on YOLO-V4 is designed.The main research contents are as follows:The current defects in transmission lines are studied.According to the fault characteristics,they are specifically divided into 23 types of targets,including 4 normal targets and 19 defect targets.Due to the lack of defect target images,the sample images are expanded by the corresponding image preprocessing form,and the corresponding data sets are made for training and verification.The application of two deep learning algorithms,YOLO-V3 and YOLO-V4,in the field of transmission line insulator detection is studied,and their advantages and disadvantages are analyzed.Compared with the traditional YOLO-V3 algorithm,the combination of the spatial pyramid pooling module and the backbone network CSPDarknet can improve the perception field of the convolution core and further enhance the learning ability of the neural network,which can not only reduce the computational load,but also ensure the accuracy of the insulator fault detection.At the same time,the feature pyramid network as the core further adds the module Enhanced Network,which shortens the path of fusion between features.In order to better see the number of various fault targets,the target counting function is added to it,which can achieve real-time defect counting.The Deep Sort algorithm is used for target tracking of transmission line faults in order to solve the problem of continuous moving target missed detection that often occurs during the process of target detection by target detection algorithm.Based on target detection,a target tracking module with Kalman filter algorithm as the core is added to track all detected targets continuously for a specific period of time.The detection edge box and the tracking edge box are obtained,and the Hungarian algorithm is used to complete the best matching.The problems that often occur in the detection of YOLO-V4 algorithm are solved.According to the practical application requirements of transmission line UAV inspection,a multi-target tracking system for power grid based on YOLO-Deep Sort is designed and implemented.The experiment shows that the system can meet the practical application requirements stably and efficiently.There are 32 figures,6 tables and 64 references in this paper.