YOLOv8-ORSDCV: a quantitative detection method for the defect area of insulators used in overhead contact systems

Abstract The replacement of insulators for overhead contact systems (OCSs) is currently mainly carried out using a combination of machine detection and manual judgment, lacking quantitative analysis of the extent of the defect. To achieve quantitative analysis of the extent of insulator defects, this paper proposes a YOLOv8-ORSDCV algorithm. The proposed algorithm employs an oriented bounding box to effectively address the tilting of the detection object caused by variations in shooting angles. To solve the problem of missing key information in nighttime insulator detection due to extremely insufficient background light, the algorithm introduces the Content-aware Reassembly of FEature (CARAFE) module; the detection accuracy of the model for targets is effectively improved. To enhance the feature extraction ability of the model for insulators and their defect details, the feature pyramid shared conv module and C2f_diverse branch block module are constructed to improve the model’s understanding of and ability to analyze image content, as well as detection speed. Finally, the OpenCV module is introduced to achieve precise quantification of the defect contour area through filtering, edge detection, and contour extraction. The research results show that, compared to the YOLOv8n algorithm, the four performance indicators (mAP@0.5, mAP@0.5–0.95, insulator detection accuracy, and insulator defect detection accuracy) of the YOLOv8-ORSDCV algorithm have been improved by 2.3%, 11.1%, 1.6%, and 3%, respectively. In quantitative detection, the YOLOv8-ORSDCV algorithm achieves an insulation error of approximately 4.43% and a defect error of about 1.5%. Due to accurately locating the location of the insulator defect, a real-time quantitative display of the defect area is achieved. The research results of this paper provide technical support for the intelligent monitoring and maintenance of high-speed railway OCSs.