Research On Insulator Fault Detection Algorithm Based On Improved YOLOv4

There are many branches of transmission lines in the oilfield power grid and the power grid structure is complex.In the complex power grid,thousands of insulators play an important role in insulation and support.They are exposed outdoors for a long time and face challenges such as lightning and pollution at any time.Therefore,insulators must be one of the key objects in the inspection of power lines,In the actual production,the operation state detection of insulators in time can prevent the occurrence of power grid accidents.During UAV patrol inspection,the insulator image is captured by the airborne camera,and then transmitted to the server through the communication network.Finally,the insulator fault diagnosis is carried out by using the target recognition algorithm.Compared with traditional manual patrol inspection,UAV patrol inspection can reduce a lot of manpower and greatly improve the efficiency of patrol inspection.However,the scope of UAV patrol inspection is large and there are many aerial images.Real time transmission and processing of images taken by airborne cameras is a huge challenge for the communication network.If real-time intelligent detection of images taken by UAV can be carried out on the airborne or mobile terminal,faults can be found as soon as possible to avoid major accidents.To solve these problems,firstly,a set of insulator data in PASCAL VOC format for UAV inspection is established,and Label Img is used to annotate the image.Secondly,the depth-separable convolution is embedded in the YOLOv4 target detection network architecture,which changes PANet network and reduces the number of network parameters.Mobile Netv2,Mobile Netv3 and Ghost Net network are introduced into the YOLOV4 network to speed up network detection while reducing network parameters.Finally,three improved algorithms and YOLOv4 algorithms are modeled and simulated by using Python language in Pycharm and other development tools,among which Ghost Net-YOLOv4 algorithm has the best comprehensive performance,the Precision of testing the fault insulator image is 98.37%and the MAP is 91.64%,the FPS of the Algorithm is about 50 times that of the original YOLO V4 algorithm and the number of parameters of the algorithm is only 17.8% of that of YOLOv4 algorithm.The Ghost Net-YOLOv4 Algorithm can detect the image taken by the UAV in real-time with thigh precision.