Research on an Insulator Defect Detection Algorithm Based on an Improved YOLOv11n Approach

Abstract

Insulator defect detection in transmission lines suffers from complex background interference and insufficient small-target recognition accuracy. To address these challenges, this paper proposes an SP-YOLO insulator defect detection algorithm based on an improved YOLOv11n model. The algorithm improves upon YOLOv11n as the baseline framework. A CP-Block module, composed of the CBAM attention mechanism and pinwheel convolution, is embedded into the backbone to enhance small-target recognition and background anti-interference capability; a GOLD-YOLO-APF neck network is designed to improve cross-layer feature fusion and transmission efficiency; and a Focaler–ShapeIoU loss function is proposed, which integrates the Focal concept with ShapeIoU to focus on hard samples while optimizing target-contour perception and bounding-box regression accuracy. Experimental results show that each optimization sub-module improves the overall model performance. After integrating all modules, the model achieves an mAP@0.5 of 86.1% and an mAP@0.5:0.95 of 55.1%, a clear improvement over the original YOLOv11n. The proposed approach attains the highest mAP@0.5 and recall among the compared methods while achieving competitive precision, with only 6.25 M parameters and 68.9 FPS, achieving a favorable balance between accuracy and lightweight design.