Study On The Detection Of Aerial Insulator Drop Under YOLOv4 Algorithm

Insulators are important components in HV transmission lines,which are not only in large number and variety,but also widely distributed in HV transmission lines.Insulator string dropping defects will have a great impact on the stable operation of the transmission network.Traditional manual inspection is affected by various aspects and has many disadvantages.The problem of intellectualized inspection is solved by studying insulator target location and insulator string dropping defect detection in aerial images.This paper firstly compares and analyzes the principles and model characteristics of common classical convolutional neural algorithm and deep learning algorithm,and determines the YOLOv4 algorithm with outstanding detection speed and accuracy as the basic algorithm for subsequent research.The main work of the paper is as follows:(1)In view of the problem of too few data samples of aerial insulator images,random cropping,horizontal flipping,color conversion and other data enhancement methods were used to expand the sample data.The expanded sample data were annotated by Labellmg,an open source tool,and the insulator data set needed for subsequent research was built.(2)for aerial insulator image size is differ,the characteristics of complicated background,target detection in the insulator on YOLOv4 network model was improved,using the model parameter tuning method improved the training speed,and selects the Mix Up data enhanced with Mosaic enhancement method can enhance the model generalization ability under different backgrounds,The accuracy of insulator target positioning and detection is improved.(3)off for insulator string of testing the characteristics of single insulator target small easy to leak,off on insulator string of defect detection on YOLOv4 network model was improved,with the method of average pooling instead of pooling biggest raised to characteristic information sequence of exploitation degree,at the same time using GBCE loss function instead of the original CBE function in model,It can improve the perception ability of small targets and reduce the missed detection rate of insulator string dropping defect detection.(4)According to the intelligent demand of inspection,the integrated detection of insulator target and string dropping defect is realized by using the joint detection of the improved insulator target location network and the insulator string dropping defect detection network.