Laser & Optoelectronics Progress, Volume. 60, Issue 4, 0415009(2023)

Steel-Plate Surface-Defect Detection Algorithm Based on Improved YOLOv5s

Yan Zhou1、*, Jiangnan Meng1、**, Jia Wu1,1、">, Zhi Luo2,2、">, and Dongli Wang1,1、">
Author Affiliations
  • 1School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, Hunan, China
  • 2Hunan Valin Xiangtan Iron and Steel Co., Ltd., Xiangtan 411105, Hunan, China
  • show less
    Figures & Tables(11)
    YOLOv5s network structure
    Mosaic data expansion with MixUp
    CBAM structure
    Fusion attention module
    AP statistics of various types of defects
    AP comparison of various types of defects before and after improvement of loss function
    Data set and statistic of samples of each defect
    Comparison of detection results before and after improvement. (a) Original YOLOv5s; (b) improved YOLOv5s
    • Table 1. Anchor box statistics table

      View table

      Table 1. Anchor box statistics table

      MethodAnchor boxFitness /%
      Raw(10,13)(16,30)(33,23)(30,61)(62,45)(59,119)(116,90)(156,198)(373,326)68.46
      Euclidean(31,42)(36,86)(80,63)(38,206)(64,136)(135,79)(199,60)(109,200)(201,207)72.87
      IoU(20,39)(25,82)(50,45)(62,82)(192,31)(34,190)(146,74)(78,166)(185,188)75.56
      Genetic(20,42)(25,82)(51,46)(176,29)(64,85)(34,190)(146,74)(78,166)(185,188)75.91
    • Table 2. Ablation experiments of YOLOv5s

      View table

      Table 2. Ablation experiments of YOLOv5s

      MethodAnchorMixUpCBAM

      Focal

      loss

      Precision /%Recall /%Number of parametersmAP /%
      YOLOv5s77.172.5703581175.4
      Improved 175.973.8703581176.2
      Improved 277.072.6703581176.9
      Improved 377.972.3706867778.2
      Improved 478.073.2706867778.4
    • Table 3. Performance comparison of different models

      View table

      Table 3. Performance comparison of different models

      NetworkSpeed /(frame·s-1mAP /%AP /%
      crazinginclusionpatchespitted_surfacerolled-in_scalescratches
      YOLOv32969.144.760.884.474.561.187.2
      DDN(ResNet34)16<2074.84875.987.478.368.490.8
      DDN(ResNet50)16<1082.362.484.790.389.776.390.1
      Faster R-CNN(ResNet34)16<2070.246.761.382.876.570.783.4
      YOLOv5s4175.437.684.292.183.761.893.1
      Improved YOLOv5s4078.448.883.891.680.572.792.8
    Tools

    Get Citation

    Copy Citation Text

    Yan Zhou, Jiangnan Meng, Jia Wu, Zhi Luo, Dongli Wang. Steel-Plate Surface-Defect Detection Algorithm Based on Improved YOLOv5s[J]. Laser & Optoelectronics Progress, 2023, 60(4): 0415009

    Download Citation

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

    Category: Machine Vision

    Received: Dec. 21, 2021

    Accepted: Mar. 29, 2022

    Published Online: Feb. 14, 2023

    The Author Email: Zhou Yan (yanzhou@xtu.edu.cn), Meng Jiangnan (mjnshizhu@163.com)

    DOI:10.3788/LOP213302

    Topics