Glass Insulator Fault Identification Method Based on Improved YOLOv5

Abstract

At present, the fault identification method of glass insulators has the problems of difficult feature extraction and poor generalization ability of the model, which leads to the low accuracy of fault identification of glass insulators. Based on the Yolov5 network, this paper introduces a lightweight general sampling operator CARAFE to solve the problem of difficult feature extraction. At the same time, the attention mechanism module SENet is added to give different channels different weights to improve recognition accuracy. In addition, this paper makes further improvements in the network structure to make the network fit the above improvements. The experimental results show that the fault recognition rate of glass insulators is significantly improved compared with the unimproved network.