A lightweight insulator defect detection algorithm based on the improved YOLOv5

Abstract

Unmanned aerial vehicle （UAV） inspections now have emerged as a predominant approach for the examination of transmission lines， with a pivotal focus on the detection of insulator defects. In this context， a lightweight insulator defect detection algorithm， founded on the improved YOLOv5， is introduced. Firstly， the lightweight Ghost convolution is employed to replace conventional convolution. Subsequently， the original feature extraction network is replaced with a repeated weighted bidirectional feature pyramid network （BiFPN） to bolster feature extraction capability across various scales. Finally， the coordinate attention （CA） mechanism is introduced to enhance the efficiency of backbone feature extraction. Experimental results reveal a significant 1.7% enhancement in the average precision of insulator detection， accompanied by a 13.1% reduction in the model’s size. This refined algorithm model not only elevates detection accuracy but also streamlines its footprint， thereby enabling more efficient and rapid insulator defect detection.

Keywords

• transmission line
• insulator
• defect detection
• yolov5