An Improved Model Based on Deep Learning for Detecting Insulator Defects

Abstract

Insulators are a vital portion of high-voltage transmission lines. Defective insulators will cause the power system to fail, resulting in significant losses. Although the traditional methods of detecting the insulator defect based on image detection have improved detection efficiency, the problem of low detection accuracy and poor real-time performance arises when the aerial image resolution is high and the environmental background is complex. In this paper, an improved YOLOv5 model is proposed to solve these issues. On the basis of the YOLOv5 model, BiFPN is used as the neck part to strengthen the feature extraction capability of the model to enhance detection accuracy. We also integrate Transformer and Coordinate Attention to capture important insulator edge information and reduce the interference of complex backgrounds. The model was tested several times on a dataset that we built ourselves. The AP value of the improved YOLOv5 model for insulator defect detection is 98.6%, and the average processing speed per image is 4.5ms. The experimental results indicate that the improved YOLOv5 model is more effective than the existing insulator defect detection methods.