Research on Transmission Line Insulator Defect Recognition Algorithm Based on Improved YOLOv8 Model

Aiming at the problems of low accuracy, high missed detection rate, and poor adaptability to complex backgrounds in insulator defect detection during UAV inspection of transmission lines, this paper proposes an improved YOLOv8 model for insulator defect recognition. Firstly, a PCFocalNeXtC2f module enhances small-target feature extraction via fused local-global interactions. Secondly, AFPN optimizes Neck feature fusion to avoid information loss. Thirdly, PIOU and ATFL losses improve positioning and anti-interference. Finally, GhostConv reduces parameters for edge deployment. Rameters, meeting the requirements of edge deployment. Experimental results demonstrate that the improved model achieves an mAP@0.5 of 96.8% on the self-constructed insulator defect dataset, which is 3.8% higher than the original YOLOv8, with a 23.9% reduction in parameter count. This verifies its effectiveness and superiority in insulator defect detection.