Surface wear detection method of fixed wire rope based on improved YOLOv8-seg

Abstract. Wear detection on the surface of fixed wire ropes is crucial for ensuring safety and improving the operational efficiency of special equipment. However, due to the irregular shape of the wear area and the complex background environment, traditional detection methods and some automation technologies are difficult to provide high-accuracy detection results. Aiming to solve this problem, we propose a wire rope surface wear detection method based on an improved YOLOv8-seg algorithm. By introducing an alterable kernel convolution to replace the traditional convolution to capture irregular features adaptively, introducing a mixed model that combines the advantages of both self-attention and convolution to enhance feature recognition under complicated environmental conditions, and introducing a multiscale convolutional attention module to improve the recognition ability of strip features, this method improves detection accuracy significantly. The experimental results show that the improved YOLOv8-seg model achieves 96.4%, 95%, and 97.5% of detection precision, recall, and mAP for wear and tear on the custom dataset at the mask level, respectively, and the precision, recall, and mAP are improved by 3.1%, 5.1%, and 2.9%, respectively, compared with the original model. Therefore, it is justified that the method effectively mitigates the insufficient precision and robustness of steel cable surface wear detection.