Abstract: The catenary cantilever insulator is a crucial component for ensuring the power supply safety of electrified railways, and its surface defects have a significant impact on the insulation performance. The "4C System" established in the railway industry can capture catenary images. However, due to problems such as complex image backgrounds, uneven illumination, small pixels of insulator surface defects, and missing feature information, it is still unable to accurately and automatically identify insulator surface defects from catenary images.For this reason, a two-stage insulator defect detection method based on YOLO v8 is proposed. In the first stage, an insulator detection model is established to detect and segment insulators from catenary images. After normalization and correction, a data set is established to train the surface defect recognition model in the second stage, which can identify five typical defects. The defect coordinates are then restored to the original image through feature symbol matching and homography matrices. The results of the verification tests carried out show that the proposed improved YOLO v8 two-stage cascaded insulator surface defect detection method can achieve an average detection precision rate of 92.5%, and all evaluation indicators are superior to existing methods.