Scholarlink

<p>To address challenges such as significant variation in defect sizes, complex background interference, and small target defects leading to low detection accuracy, slow speed, and inaccurate localization in steel wire rope inspection, we propose the REL-YOLO model&#xa0;based on YOLOv8. The model uses the RepNCSPELAN4 module instead of the original C2f module in the backbone network. This enriches the feature information of the extracted targets and ensures a balance between detection speed and accuracy. E-FPN is proposed to be applied to the neck network to enhance the spatial interaction capability and improve the localization of small targets. The LSDECD lightweight detection head is proposed, which can not only reduce the interference of a complex background and improve the positioning accuracy but also greatly reduce the number of head parameters and speed up the convergence of the model. The experimental results show that the improved model achieves an mAP50 of 0.94, a number of parameters of 1.68M and a GFLOPs of 4.5, where the mAP50 algorithm is 3% higher than the original model. Therefore, REL-YOLO exhibits superior performance in detecting defects in elevator wire ropes, enabling fast and accurate detection of defects in these 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

REL-YOLO: surface defect detection of elevator wire rope based on improved YOLOv8

  • 期刊论文

  • 发表日期：

    2025年7月5日

查看全文
  • Xiaotian Lv,
  • Yuan Lan,
  • Bing Wu,
  • Linkai Niu,
  • Lina Li,
  • Wangzhe Du

摘要

To address challenges such as significant variation in defect sizes, complex background interference, and small target defects leading to low detection accuracy, slow speed, and inaccurate localization in steel wire rope inspection, we propose the REL-YOLO model based on YOLOv8. The model uses the RepNCSPELAN4 module instead of the original C2f module in the backbone network. This enriches the feature information of the extracted targets and ensures a balance between detection speed and accuracy. E-FPN is proposed to be applied to the neck network to enhance the spatial interaction capability and improve the localization of small targets. The LSDECD lightweight detection head is proposed, which can not only reduce the interference of a complex background and improve the positioning accuracy but also greatly reduce the number of head parameters and speed up the convergence of the model. The experimental results show that the improved model achieves an mAP50 of 0.94, a number of parameters of 1.68M and a GFLOPs of 4.5, where the mAP50 algorithm is 3% higher than the original model. Therefore, REL-YOLO exhibits superior performance in detecting defects in elevator wire ropes, enabling fast and accurate detection of defects in these wire ropes.