Defect Detection Of Composite Insulators For Transmission Lines Based On Deep Learning

Electricity is a kind of renewable resource that is both clean and environmentally friendly.It is closely related to our industrial production and daily life.It is also the main way to reduce air pollution and improve energy efficiency.Composite insulators widely used in UHV transmission lines are an important link to ensure the safe and stable transmission of electrical energy resources.In this paper,research is conducted on the defect detection of composite insulators.The data used in this article is a set of pictures of transmission line towers taken by the drone.In order to accurately locate the composite insulator image and the position of the composite insulator,this paper firstly uses the improved Faster-RCNN model to extract the composite insulator image that occupies only a small part of the entire picture.According to the characteristics of composite insulators,combined with the actual application scenarios,the detection effect of composite insulators is improved by changing the prediction frame and improving the loss function and evaluation standards.For the composite insulator image detected by the improved Faster-RCNN model,data enhancement is performed to a certain extent to improve the data imbalance and the completeness of the data set through various image processing,combined with the actual external conditions during drone shooting.In this paper,I have made a composite insulator data set,and these data will be public for future researchers.The defect of the composite insulator is mainly the edge deformation.This defect may cause the umbrella skirt of the composite insulator to be cracked or broken,resulting in an uneven voltage distribution on the composite insulator.At the same time,the deformed part is more likely to produce pollution accumulation,which reduces the insulation performance Pollution flash.Therefore,paying attention to the condition of the edge of the composite insulator can effectively judge the state of the composite insulator,and plays a key role in the defect detection of the composite insulator.The edge-supervised classification network designed in this paper is added to the edge detection module.The edge detection map generated by the edge detector of the image edge detection map generated by the edge detector supervised by the feature map allows the entire feature extraction network to learn the image edge features well.These features that incorporate more image edge information are used for subsequent image classification to obtain better results.Through the two-stage operation of target detection and image classification,the composite insulator in the original image was successfully classified into normal or defect categories,and the defect detection of the composite insulator was realized.