In order to realize the real-time monitoring of the surface damage of mine hoist wire rope and improve the robustness and detection efficiency of defect recognition in industrial scene, a wire rope damage identification method based on machine vision is constructed. By building an industrial camera device to collect the image data of the wire rope surface in operation, the single-stage target detection framework YOLO v5 s network structure is improved for the small damage characteristics, the multi-scale feature fusion mechanism and the bounding box loss function are optimized, and the transfer learning strategy is used to optimize the initialization process of the network parameters, so as to improve the generalization ability of the model under the condition of limited samples. The experimental data show that the average accuracy of the improved algorithm on the test set reaches 89%, which is 4% higher than the benchmark model, and 19%, 16% and 10% higher than the SSD, Faster R-CNN and YOLOv3 benchmark architectures, respectively. The technical advantages of the method in wire rope damage detection are verified.
