A Composite Insulator Overheating Defect Detection System Based on Infrared Image Object Detection

Abstract

Composite insulators are important components of power transmission lines and their quantity is huge, and their overheating defects can cause serious power accidents, so it is crucial to detect the overheating defects of composite insulators on a regular basis. Current detection methods suffer problems such as low detection accuracy and efficiency. To solve these problems, we propose a method for detecting overheating defects of composite insulators based on infrared images and computer vision. The system is divided into two parts: composite insulator detection and key point extraction. During the experiment, we found that there are problems of sample imbalance and inaccurate positioning of prediction box, so we proposed Equalized Fully Convolutional One-Stage object detection (FCOS), which adds the sample equalization strategy and multi-dimensional dynamic attention mechanism on the basis of FCOS, and improves the Centerness and proposes the center positioning confidence. The results show that the performance of Equalized FCOS is significantly improved compared with FCOS, with mAP, AP50, AP75, APm, and APl improved by 7.6%, 5.1%, 12.1%, 4.6%, and 9.7% respectively. Finally, considering the practical application requirements, we extracted the key points of the mandrels and performed temperature analysis. We have realized the automated inspection of composite insulators.