Research on Lightweight Insulator Defect Detection Algorithm for Improved YOLOv5s

Abstract

To address the issues of missed detections and low detection accuracy in insulator defect detection tasks using drones, and to meet the real-time detection requirements of drones, an improved insulator defect detection algorithm based on YOLOv5s is proposed,better preserve image details and prevent information loss that could lead to missed detections. Secondly, a lightweight Group Collaborative Attention (GCA) mechanism is proposed and integrated into the backbone network to enhance the model's ability to extract features of small targets without significantly increasing the number of parameters. Additionally, a Pyramid bottleneck structure is designed and integrated into the residual module to increase the receptive field and prevent the loss of edge feature information during iterations. Finally, channel pruning and fine-tuning are applied to the improved model to further reduce the number of parameters while preventing excessive loss of accuracy. On the Chinese Power Line Insulator Dataset (CPLID), this algorithm achieves 12.2%, 10.0%, 7.4%, 13.3%, 2.5%, and 2.3% higher accuracy compared to YOLOv3-tiny, YOLOv4-tiny, YOLOv5s, YOLOv7-tiny, YOLOv8s, and YOLOv10s respectively. The detection time per image is reduced from 7.4ms to 5.4ms, achieving an FPS of 147.3. The total number of model parameters is reduced by 37.9%. Heatmap and visualization comparative analysis show that the improved model effectively addresses the issues of missed detections and low detection accuracy.