Scholarlink

<p>Steel wire rope is one of the most common components in industry, and its damage degree has a great impact on personal safety. Therefore, it is crucial to detect the early surface damage of steel wire rope. This paper introduces the semantic segmentation algorithm into deep learning to the surface damage detection of steel wire rope for the first time and proposes a steel wire rope broken wire damage segmentation method WRU-Net. First, lightweight transformation of the backbone network in U-Net is carried out to reduce the parameters of the model and improve the detection speed. Then, to improve the detection accuracy when facing broken wires with differences, a separable spatial attention pyramid module is proposed to modify the decoder. The feature alignment fusion module is then used to align and fuse the high-resolution feature map and the high semantic feature map to enhance the semantic information of the predicted map, thereby further improving performance. WRU-Net and different semantic segmentation models were tested on the wire rope surface damage dataset. The results show that WRU-Net has the best results, with the mean intersection over union (mIoU) of 88.14%, and the model size is also the smallest at 22.38&#xa0;MB. Therefore, this method provides a new solution for surface damage detection of wire ropes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detection of Surface Damage on Steel Wire Ropes Based on Improved U-Net

  • 期刊论文

  • 发表日期：

    2025年2月4日

查看全文
  • Jilin Wei,
  • Juwei Zhang,
  • Hongli Wang

摘要

Steel wire rope is one of the most common components in industry, and its damage degree has a great impact on personal safety. Therefore, it is crucial to detect the early surface damage of steel wire rope. This paper introduces the semantic segmentation algorithm into deep learning to the surface damage detection of steel wire rope for the first time and proposes a steel wire rope broken wire damage segmentation method WRU-Net. First, lightweight transformation of the backbone network in U-Net is carried out to reduce the parameters of the model and improve the detection speed. Then, to improve the detection accuracy when facing broken wires with differences, a separable spatial attention pyramid module is proposed to modify the decoder. The feature alignment fusion module is then used to align and fuse the high-resolution feature map and the high semantic feature map to enhance the semantic information of the predicted map, thereby further improving performance. WRU-Net and different semantic segmentation models were tested on the wire rope surface damage dataset. The results show that WRU-Net has the best results, with the mean intersection over union (mIoU) of 88.14%, and the model size is also the smallest at 22.38 MB. Therefore, this method provides a new solution for surface damage detection of wire ropes.