Opto-Electronic Engineering, Volume. 51, Issue 5, 240028(2024)

Defects detection for cable surface of cable-stayed bridge based on improved YOLOv5s network

Pengfeng Wang1... Yuntang Li2,*, Yongyong Huang2, Wenkai Zhu2, Jie Lin2 and Binrui Wang2 |Show fewer author(s)
Author Affiliations
  • 1College of Modern Science and Technology, China Jiliang University, Jinhua, Zhejiang 322002, China
  • 2College of Mechanical and Electrical Engineering, China Jiliang University, Hangzhou, Zhejiang 310018, China
  • show less
    Figures & Tables(19)
    Structure of the conventional YOLOv5s network
    SPPF module
    Schematic diagram of the prediction box
    Structure of the improved YOLOv5s network
    TRANS module
    Multi-head attention
    GhostBottleneck module
    Ghost structure
    Angle loss
    Distance loss
    Data collection of cable surface defects
    Mosaic data augmentation
    Annotation instance
    Loss value variation curve
    Comparison of different network detection results
    • Table 1. Prior box size of conventional YOLOv5s

      View table
      View in Article

      Table 1. Prior box size of conventional YOLOv5s

      特征图尺寸先验框尺寸
      特征图180×80 (10, 13), (16, 30), (33, 23)
      特征图240×40 (30, 61), (62, 45), (59, 119)
      特征图320×20 (116, 90), (156, 198), (373, 326)
    • Table 2. Ablation comparative experiment

      View table
      View in Article

      Table 2. Ablation comparative experiment

      算法AP /%mAP /%FPS
      abcd
      常规YOLOv5s90.4289.0883.7791.5787.7156
      方法191.9292.5494.0196.1393.6551
      方法294.0991.8190.3295.6292.9664
      方法393.2092.2391.2695.8393.1368
      方法495.1191.2993.4297.2294.2668
    • Table 3. Detection results of different networks on the surface defect dataset of the cable

      View table
      View in Article

      Table 3. Detection results of different networks on the surface defect dataset of the cable

      网络mAP /%FPS
      Faster R-CNN90.573
      文献[12]87.8916
      YOLOv489.3720
      常规YOLOv5s87.7159
      常规YOLOv8s92.3851
      改进YOLOv5s94.2668
    • Table 4. Detection results of different networks on VOC 2007 dataset

      View table
      View in Article

      Table 4. Detection results of different networks on VOC 2007 dataset

      网络mAP /%FPS
      Faster R-CNN76.384
      文献[12]71.9121
      YOLOv473.1226
      常规YOLOv5s72.5763
      常规YOLOv8s76.8257
      改进YOLOv5s78.2171
    Tools

    Get Citation

    Copy Citation Text

    Pengfeng Wang, Yuntang Li, Yongyong Huang, Wenkai Zhu, Jie Lin, Binrui Wang. Defects detection for cable surface of cable-stayed bridge based on improved YOLOv5s network[J]. Opto-Electronic Engineering, 2024, 51(5): 240028

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Article

    Received: Jan. 26, 2024

    Accepted: Apr. 24, 2024

    Published Online: Jul. 31, 2024

    The Author Email: Li Yuntang (李运堂)

    DOI:10.12086/oee.2024.240028

    Topics