Research on Improving Method of RCT-YOLOv5s Network for Surface Defect Extraction of Steel Wire Rope

According to the particularity of steel wire rope surface defects., several YOLO models were analysed and then YOLOv5s was selected as the basis to improve in this paper: To enhance the capability of defect feature extraction., RepVGG was used to replace two C3 units of network Backbone; To improve the ability of multi-scale feature expression., CoTN was used to replace C3 unit of Neck; In order to improve the recognition accuracy of small defects., a small target detection head was added to the head in network. At the same time., the loss function EIoU was used to instead of CIoU to achieve the goal that the detection accuracy and the defect detection speed was improved. It was shown from the experimental results that the new RCT-YOLOv5s network can improve the accuracy of surface defect of steel wire rope, especially for tiny defects, and the recognition accuracy of the surface defect of steel wire rope was improve by 4.5% compared with that of using the basic YOLOv5s.