Research on Transmission Line Hardware Identification Based on Improved YOLOv5 and DeblurGANv2

摘要

In order to accurately identify the high voltage transmission line fittings and guide the line patrol robot to achieve the corresponding obstacle-crossing action according to different fittings, a real-time detection network based on deep learning DeblurGANv2 and YOLOv5 target detection algorithm was proposed. In view of the problems of large network model, large amount of computation and low operation efficiency of YOLOv5 algorithm, the lightweight improvement of YOLOv5 algorithm was carried out to improve the detection speed of the algorithm. Shufflenetv2 was used to replace CSPDarknet-53 in the original network, and ECA module was introduced into Shufflenetv2.Enhance the model’s focus on valid features. At the same time, DeblurGANv2 super-resolution reconstruction algorithm is introduced to deblur the image of the fittings to generate high-quality images and improve the recognition accuracy. Experimental results based on homemade datasets show that,compared with the original YOLOv5 model, the parameter amount of the proposed method is reduced by 50.2%, the weight file size is only 5.3M, a reduction of 62.7%, and mAP@ (0.5) is increased by 1.3%. The proposed method can successfully detect the high voltage transmission line fittings and ensure the accuracy of fuzzy target detection.