Chinese Journal of Lasers, Volume. 51, Issue 20, 2002102(2024)

Aluminum Alloy Weld DR Image Defect Detection Technology Based on YOLOv7TS

Lei Wu1, Yukun Chu1, Honggang Yang1, and Yunxia Chen2、*
Author Affiliations
  • 1School of Mechanical Engineering, Shanghai Dianji University, Shanghai 201306, China
  • 2Shanghai Polytechnic University, Shanghai 201209, China
  • show less
    Figures & Tables(13)
    YOLOv7Tiny frame diagram
    Structure of each module in YOLOv7Tiny. (a) Structure diagram of CBL module; (b) structure diagram of ELAN module;
    Structure diagram of improved YOLOv7Tiny
    Structure diagram of S-ELAN module
    Data annotation process
    Defect images after data augmentation. (a) Original defect image; (b) after horizontal rotation; (c) after vertical flipping; (d) after randomly adjusting image brightness; (e) after adding salt and pepper noise; (f) after adding Gaussian noise
    Mosaic data augmentation
    Comparison of models detection effects
    Comparison of detection effect for slag inclusion defects in small target types. (a) Detection effect of the original model; (b) detection effect of the improved model
    Comparison of detection effect for three types of defects. (a) Detection effect of the original model; (b) detection effect of the improved model
    • Table 1. Comparison of precision indexes of different models

      View table

      Table 1. Comparison of precision indexes of different models

      ModelAP /%mAP@0.5 /%
      PoreSlag inclusionIncomplete penetration
      YOLOv7TS89.994.296.393.3
      YOLOv7Tiny81.790.594.188.7
      YOLOv5s88.291.177.185.5
      YOLOv4Tiny63.977.645.662.3
      SSD83914974.11
      DETR70852660.33
    • Table 2. Comparison of parameter number, weight file size and inference speed of different models

      View table

      Table 2. Comparison of parameter number, weight file size and inference speed of different models

      ModelParameter numberWeight file size /MBFPS
      YOLOv7TS571920911.8208
      YOLOv7Tiny602040012.3222
      YOLOv5s706893613.7156
    • Table 3. Ablation experimental results

      View table

      Table 3. Ablation experimental results

      No.TSCODECARAFESPD-ConvLightweightSimAMmAP@0.5 /%Parameter number
      188.75757232
      2926020400
      392.611075209
      490.75719209
      593.35719209
    Tools

    Get Citation

    Copy Citation Text

    Lei Wu, Yukun Chu, Honggang Yang, Yunxia Chen. Aluminum Alloy Weld DR Image Defect Detection Technology Based on YOLOv7TS[J]. Chinese Journal of Lasers, 2024, 51(20): 2002102

    Download Citation

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

    Category: Laser Forming Manufacturing

    Received: Oct. 20, 2023

    Accepted: Jan. 8, 2024

    Published Online: Oct. 11, 2024

    The Author Email: Chen Yunxia (cyx1978@yeah.net)

    DOI:10.3788/CJL231313

    CSTR:32183.14.CJL231313

    Topics