Laser & Optoelectronics Progress, Volume. 60, Issue 22, 2215005(2023)

Gear Surface Defect Detection Method Based on Improved YOLOx Network

Shuwen Zhang1,2, Zhenyu Zhong1,2, and Dahu Zhu1,2、*
Author Affiliations
  • 1Hubei Key Laboratory of Advanced Technology for Automotive Components, School of Automotive Engineering, Wuhan University of Technology, Wuhan 430070, Hubei, China
  • 2Hubei Collaborative Innovation Center for Automotive Components Technology, School of Automotive Engineering, Wuhan University of Technology, Wuhan 430070, Hubei, China
  • show less
    Figures & Tables(17)
    The main structure of the YOLOx algorithm
    Structure diagram of CSP_X
    Structure of decoupled head
    Schematic diagram of the structure of classic ASFF
    Revised structural diagram
    Schematic diagram of SE-Layer
    Flowchart of the SE-Layer
    Schematic diagram of ECA
    Structure of improved YOLOx
    Change curves of loss function value during original YOLOx training
    Change curves of the loss function value during the training of the improved YOLOx
    Comparison of the mAP@0.5 curves by the improved and original YOLOx network
    Comparison of the metallic gear defect detection results by the original and improved YOLOx networks. (a) Original YOLOx;(b) improved YOLOx
    Detection site of gear surface defects. (a) The first detection station; (b) the second detection station
    • Table 1. Comparison table of ablation experiments

      View table

      Table 1. Comparison table of ablation experiments

      ModuleRecallPrecisionmAP@0.5
      YOLOx75.178.374.2
      YOLOx+ASFF70.187.376.1
      YOLOx+ECA78.784.277.7
      YOLOx+Varifocal loss76.785.877.4
      YOLOx+ASFF+ECA78.186.680.7
      YOLOx+ECA+Varifocal loss75.686.178.0
      YOLOx+ASFF+Varifocal loss75.685.676.6
      YOLOx+ASFF+Varifocal loss+ECA81.282.983.6
    • Table 2. Comparison of different algorithms

      View table

      Table 2. Comparison of different algorithms

      AlgorithmRecall /%Precision /%mAP@0.5 /%FPS
      YOLOx75.178.374.245
      YOLOv5s63.693.374.333
      YOLOv5m72.788.976.719
      SSD70.291.675.813
      Faster R-CNN77.571.777.115
      Improved YOLOx81.282.983.634
    • Table 3. Operating conditions of the detection system

      View table

      Table 3. Operating conditions of the detection system

      DataTotal numberMissing quantityReturned quantityMisjudgmentAccuracy /%
      2022-06-284960122651695.0
      2022-06-295700152351395.5
      2022-06-30400010130895.3
      2022-07-01500072101594.6
      2022-07-026000121831595.3
      2022-07-035500182721894.4
      2022-07-047000263461995.6
    Tools

    Get Citation

    Copy Citation Text

    Shuwen Zhang, Zhenyu Zhong, Dahu Zhu. Gear Surface Defect Detection Method Based on Improved YOLOx Network[J]. Laser & Optoelectronics Progress, 2023, 60(22): 2215005

    Download Citation

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

    Category: Machine Vision

    Received: Jan. 9, 2023

    Accepted: Mar. 6, 2023

    Published Online: Nov. 16, 2023

    The Author Email: Zhu Dahu (dhzhu@whut.edu.cn)

    DOI:10.3788/LOP230469

    Topics