Research On Insulator Defect Detection Of Transmission Line Based On Deep Learning

With the rapid development of power system and the increasing demand for power,the scale of transmission lines in China is expanding sharply.It is very important to regularly check the electrical equipment involved to ensure the reliability of power supply.Insulators are common insulation devices in transmission lines.Because they work in outdoor air for a long time,the environment conditions are bad,and they are prone to breakage or loss,which poses a great threat to the safe and stable operation of power system,so it is more important to inspect and troubleshoot the insulators.At present,the inspection of insulators mainly focuses on unmanned aerial vehicle patrol.The image of insulators is obtained by aerial photography of unmanned aerial vehicle,and the related methods are used to identify,locate and detect the defects of the insulators in the image.This paper takes the insulator image as the research object,and takes the defect detection of the insulator as the target,and implements the defect detection of the insulator based on image segmentation and target detection methods respectively.The following aspects are mainly studied:(1)In order to improve the image sharpness and noise of some insulators,linear transformation and filter processing are used to enhance the image.In view of the insufficient number of insulator images,an affine transformation method is used to expand the data of the insulator images.Finally,the insulator image is labeled with the grabcut tool to get the data set of this article.(2)An image segmentation method based on VGG16 network combined with U-net network is presented to overcome the shortcomings of traditional image segmentation methods such as poor segmentation results and low segmentation accuracy.Using the similar features of VGG16 network and U-net network in feature extraction stage,the VGG16 network is improved as the feature extraction part of U-net network,and the insulator image segmentation network used in this paper is obtained.Experiments show that compared with U-net network,the improved network can significantly improve the isolation effect of the insulator.(3)A defect detection method for insulator slices based on least squares fitting is presented according to the elliptical and regularly distributed characteristics of the separated insulator strings.All the pixel points of the insulator string are traversed by using the method of mathematical modeling.The fitted straight line is scanned along the insulator string to determine if there is a defect in the insulator sheet,locate and record the location of the defect,and mark it in the original drawing.(4)An improved YOLOv5 algorithm for defect detection of insulators is presented in view of the low accuracy and slow detection speed of target detection algorithm.First,in order to improve the accuracy of target detection,attention mechanism is introduced in the backbone section of YOLOv5.Secondly,to solve the unstable regression problem of the target frame,the PIo U loss function is used instead of the original GIo U loss function.Finally,the detection results of the improved algorithm are compared and analyzed.