Broken Defect Localization of Glass Insulators with Oriented Bounding Box Model

Abstract

Glass insulators are essential for ensuring highvoltage transmission lines’ safe and efficient operation, facilitating reliable power transfer over long distances. However, accurately localizing defects in insulators that are rotated poses a challenge. To address this challenge, this paper proposes a novel approach for detecting broken insulator defects, utilizing a single-stage object detector model integrated with the concept of oriented bounding boxes. By converting annotations to the YOLO OBB format, the model can detect the defect on tilted insulators, enhancing the accuracy of defect localization. Experimental evaluations showcase the superior performance of this approach compared to the conventional YOLOv8 model. The YOLOv8n-OBB model, in particular, demonstrates a remarkable mean Average Precision (mAP@50) of 96.9%, representing a significant advancement in glass insulator defect identification. This innovative method enhances defect detection accuracy and contributes to the overall reliability and efficiency of high-voltage transmission line maintenance and operation.