An Aerial Photography-Based Insulation Defect Detection Model for Insulators, Developed using YOLOv11 and Multimodal Detection Techniques

Abstract

The rapid progress of drone technology has made intelligent aerial-image-based power equipment inspection a vital research topic for ensuring power system safety and maintenance. To overcome the high false-alarm rates of traditional insulator defect detection under complex conditions and the insufficient exploitation of multimodal information, this paper presents a detection framework built upon an enhanced YOLOv11 architecture combined with multimodal data fusion. By introducing a cross-modal cooperative perception mechanism and a mid-level fusion strategy, the proposed method enables deep interaction between visible-light and infrared features, thereby strengthening the saliency representation of defect areas. Experiments on a self-constructed dataset indicate that the model achieves an average precision of 94.57%, an accuracy of 94.12%, and a recall of 92.19%, with a detection speed of 24 FPS, satisfying the real-time requirements of power-line inspection.