Detection of Transmission Line Insulator Defect Based on Improved YOLOv10

The transmission line insulator is a critical component in the power system for wire support and ground insulation, which has impact on safety and stability of the system. The traditional insulator inspection methods, such as manual or robot methods, are mostly uneffective. The deep learning technology has promoted the development of insulator defect detection using visionassisted Unmanned Aerial Vehicle (UAV) images. In this paper we propose an improved YOLOv10 algorithm for insulator defect detection. Firstly, based on an in depth analysis of insulator defect characteristics, the traditional two-category classification of insulator defects has been innovatively refined into three insulator categories and five defect types, significantly enhancing defect judgment accuracy and precision. Addressing the current limitation that drones can only capture insulator images in fair weather conditions, this paper introduces an online data augmentation technique to simulate adverse environmental conditions. By adding random cloud, fog, and raindrop effects, as well as adjusting brightness and contrast, this approach mimics drone-captured insulator images in various harsh environments, thereby strengthening the model's feature extraction capabilities in complex backgrounds. Then, to improve spatial feature extraction in captured images within the backbone network, the EMA attention mechanism is applied to enhance the traditional MSHA attention mechanism in the PSA module, boosting the accuracy of insulator defect detection. Finally, the CGLU module is applied to refine the MLP in the additive block, resulting in the additiveblock_CGLU, which is further integrated into the C2f structure. This innovative integration improves detection speed while maintaining accuracy and reducing computational load. Experimental results indicate that, the proposed method can achieve Precision of 84.2%, Recall of 81.9%, and reaches mAP50 of 84.1%.