Optics and Precision Engineering, Volume. 32, Issue 8, 1227(2024)

Improving the lightweight VTG-YOLOv7-tiny for steel surface defect detection

Liming LIANG... Pengwei LONG*, Yao FENG and Baohe LU |Show fewer author(s)
Author Affiliations
  • School of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou341000, China
  • show less
    Figures & Tables(18)
    VTG-YOLOv7-tiny model structure
    VoVGA-FPN network structure
    GConv and SC network structure
    Gbottleneck and VoVGCSP network architecture
    TCA network structure
    GSConv network structure
    Image of various defects on the steel surface
    Comparison of AP values of this paper's algorithm and the original model for various types of defects
    Comparison of the detection effect of this paper's algorithm and the original model
    • Table 1. Improved anchoring frame sizes for each detection layer

      View table
      View in Article

      Table 1. Improved anchoring frame sizes for each detection layer

      特征图尺寸感受野NEU-DET数据集锚定框大小Severstal数据集锚定框大小
      10×10超大(361,266)(246,560)(490,540)(478,238)(204,502)(606,613)
      20×20(117,517)(217,355)(438,190)(428,122)(267,235)(112,576)
      40×40(51,529)(176,193)(524,77)(48,573)(159,187)(157,318)
      80×80(61,130)(135,111)(82,238)(60,121)(132,111)(76,239)
    • Table 2. Improved VoVGA-FPN comparison experiments

      View table
      View in Article

      Table 2. Improved VoVGA-FPN comparison experiments

      ModelmAP%Params/MFLOPs/GFPS/ms
      PANet68.76.0213.1108
      AFPN70.77.1114.1108
      Ours71.26.3512.7111
    • Table 3. Experiments with different positions of the TCA module

      View table
      View in Article

      Table 3. Experiments with different positions of the TCA module

      LocationmAP%Params/MFLOPs/G
      Baseline68.76.020 113.1
      情况170.16.020 413.2
      情况270.86.020 513.2
      情况369.86.020 813.2
    • Table 4. TCA comparison experiments

      View table
      View in Article

      Table 4. TCA comparison experiments

      ModelmAP%Params/MFLOPs/G
      Baseline68.76.020 113.1
      SE70.411.57030.8
      GC70.16.037 313.2
      CA68.36.034 513.5
      CBAM70.16.028 613.2
      TA69.96.020 513.2
      TCA(Ours)70.86.020 513.2
    • Table 5. Experiments with different positions of the GSConv module

      View table
      View in Article

      Table 5. Experiments with different positions of the GSConv module

      LocationmAP%Params/MFLOPs/GFPS/ms
      A68.76.0213.1108
      B70.14.829.994
      C70.34.3110.3121
      D72.03.127.1100
    • Table 6. New detection layer with Neck with and without convolution experiments

      View table
      View in Article

      Table 6. New detection layer with Neck with and without convolution experiments

      ModelmAP%Params/MFLOPs/GFPS/ms
      A68.76.0213.1108
      B72.17.8713.5111
      C73.010.213.980
    • Table 7. Ablation experiments on NEU-DET and Severstal datasets

      View table
      View in Article

      Table 7. Ablation experiments on NEU-DET and Severstal datasets

      DatasetMethodmAP%Params/MFLOPs/GFPS/msP%R%
      NEU-DETM168.76.0213.110861.472.7
      M270.86.0213.211963.873.7
      M372.57.8713.511366.774.2
      M473.67.0412.77667.172.0
      M574.45.418.98768.574.5
      SeverstalM165.66.0213.16158.165.2
      M267.66.0213.27965.870.2
      M371.57.8713.57064.572.1
      M472.77.0412.75266.867.7
      M574.15.418.97867.072.2
    • Table 8. Comparison of experiments before and after enrichment of the Severstal dataset

      View table
      View in Article

      Table 8. Comparison of experiments before and after enrichment of the Severstal dataset

      DatasetModelmAP%Params/MFLOPs/GFPS/ms
      SeverstalAM171.56.0213.189
      M575.25.418.994
      SeverstalBM165.66.0213.161
      M574.15.418.978
    • Table 9. Experiments comparing NEU-DET and Severstal datasets

      View table
      View in Article

      Table 9. Experiments comparing NEU-DET and Severstal datasets

      DatasetModelmAP0.5/%Params/ MFLOPs/GSize/MB
      NEU-DETFaster R-CNN65.772.0167.3108.0
      SSD61.041.1145.393.1
      YOLOv367.061.5155.0117.0
      YOLOv3-tiny46.58.6712.916.6
      YOLOv451.052.5119.8102.6
      YOLOv4-tiny54.65.9016.1822.5
      YOLOv5s70.17.0716.413.7
      YOLOX-s71.88.021.616.3
      YOLOv770.037.2104.874.8
      YOLOv7-tiny68.76.0213.111.7
      YOLOv8s72.111.128.421.4
      文献[174.123.9
      文献[1682.415.3729.7
      文献[1877.27.03
      VTG-YOLOv7-tiny(Ours)74.45.418.910.8
      SeverstalYOLOv3-tiny56.48.6712.916.6
      YOLOv4-tiny59.65.916.1822.5
      YOLOv5s72.47.0716.413.7
      YOLOX-s73.88.021.616.3
      YOLOv7-tiny65.66.0213.111.7
      YOLOv8s71.811.128.421.4
      VTG-YOLOv7-tiny(Ours)74.15.418.910.8
    Tools

    Get Citation

    Copy Citation Text

    Liming LIANG, Pengwei LONG, Yao FENG, Baohe LU. Improving the lightweight VTG-YOLOv7-tiny for steel surface defect detection[J]. Optics and Precision Engineering, 2024, 32(8): 1227

    Download Citation

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

    Category:

    Received: Oct. 16, 2023

    Accepted: --

    Published Online: May. 29, 2024

    The Author Email: LONG Pengwei (2637018663@qq.com)

    DOI:10.37188/OPE.20243208.1227

    Topics