Intelligent Identification Of Transmission Line Insulator Faults Based On Deep Learning

With the successful implementation of the "14th Five-Year Plan",the erection area of transmission lines in China is gradually expanding,and the requirements for transmission capacity and transmission distance are getting higher and higher.Insulators,as important equipment of transmission lines,are easily affected by various factors in the process of UAV inspection,and the detection results often result in leakage and misdetection.Combining deep learning detection algorithm to improve the transmission line UAV inspection system can effectively solve the current problems in detection.In this paper,taking transmission line insulators as the detection object,corresponding image pre-processing methods and broadening methods are proposed for the problems of inconspicuous image features,noise pollution and low data capacity,and insulator dataset based on COCO format is constructed by using Labelbee.For the problem of low recognition accuracy of insulator detection algorithm,the two-stage detection algorithm,single-stage detection algorithm and backbone network are studied and the experimental data are compared and analyzed.Based on this,MS-COCO pre-training strategy is proposed to improve the Cascade R-CNN algorithm by combining FPN module and ResNeXt-101 network,so as to improve the recognition accuracy of insulator detection algorithm.The image pre-processing method proposed in this paper effectively improves the image quality of the data and reasonably expands the data capacity in combination with the image augmentation technique,which significantly improves the low recognition accuracy of the insulator detection algorithm due to the low data quality and insufficient capacity.In this paper,we systematically and completely analyze the mainstream insulator detection algorithms,and verify the effectiveness of the improved Cascade R-CNN X101 model by combining the mAP value and other related evaluation indexes.Compared with the detection algorithms such as Faster R-CNN and Retina Net,this model makes a small increase in computation and number of parameters,but it is significantly better than the former in detection accuracy,and can effectively cope with the cases of false detection,missed detection and unrecognized due to special environment in line inspection.The research in this paper provides a new way of thinking for realizing the fault detection intelligence of transmission line insulators,and has certain reference value for engineering applications.