With the development of high-speed railway, catenary sling defects have become a big hidden danger that affects the driving safety. Therefore, it is extremely important to detect and troubleshoot the sling defects. A new visual inspection method for catenary sling defects is proposed in this paper, focusing on two key points: foreign body, hard bending detection and no-stress detection. With regard to the first aspect, a light YOLOv3 network is presented to extract sling regions, and then Faster R-CNN is employed to inspect foreign body and hard bending. This can acquire an attractive inspection result, due to removal of redundant areas and defect amplification in a catenary image. With regard to the second, a new method based on traditional visual called self-defined linear difference (SLDD) is proposed to inspect no-stress after extracting catenary sling area using the light YOLOv3. This solves the problem of low detection rate caused by less train sample for deep learning. The experiment shows that Faster R-CNN has a good effect on foreign body and hard bending inspection after sling areas extraction, and the precision rate is more than 90%. And further experiment also shows that the light YOLOv3 combined with SLDD can achieve a better detection result for sling no-stress.