Detection of Transmission Line Insulator Defect Based on MSD-YOLO11

Detecting insulator defects in power transmission lines is critical for the reliable operation of electrical grids. Recent advancements in deep learning and image processing have enabled the efficient identification of such defects from extensive UAV aerial imagery, positioning this approach as a promising research focus on enhancing power system stability and security, This paper presents a novel seven-class detection framework for insulator defects, compared to traditional binary classifications of insulator defects. This framework improves the accuracy of distinguishing between different materials and types of defects, enabling more focused maintenance and better fault analysis. Based on original YOLO11, we propose a novel insulator defect detection model named MSD-YOLO11. In the backbone network, the newly designed C3k2_ MultiscaleEnhance structure is introduced for multi-scale feature fusion, which addresses the limitations of the traditional C3k2 module in extracting multi-scale and fine-grained features, particularly in complex backgrounds. This enhancement significantly improves edge detection performance. Additionally, the MPDIOU (Multi-Part Dynamic Intersection Over Union) loss function is proposed, effectively boosting detection accuracy for non-overlapping targets and enhancing the robustness of the model. Finally, we propose an online data augmentation technique that simulates adverse conditions such as clouds, fog, and rain, as well as variations in brightness and contrast. This data augmentation method enables the network model to better adapt to complex backgrounds in insulator defect detection, boosting detection performance. Experimental results show that the proposed model attains a Precision of 91.7%, a Recall of 85.3%, and achieves an mAP50 of 89.7%, which has better detection performance and lower computational load, compared to traditional detection methods.