Journal of Applied Optics, Volume. 44, Issue 1, 104(2023)

Defect detection on complex texture surface based on optimized ResNet

Lixing LIN1... Zhenping XIA1,2,*, Hao XU1, Yu SONG2 and Fuyuan HU2 |Show fewer author(s)
Author Affiliations
  • 1College of Physical Science and Technology, Suzhou University of Science and Technology, Suzhou 215009, China
  • 2College of Electronics and Information Engineering, Suzhou University of Science and Technology, Suzhou 215009, China
  • show less
    Figures & Tables(13)
    Schematic diagram of method for transfer learning feature mapping
    Data set preprocessed by original images
    Comparison of four groups without and with surface defects, and red dotted line frame is location of defect
    Schematic diagram of construction method for simulation data set
    Comparison of accuracy and convergence of different activation functions on test data sets
    Comparison of accuracy and convergence of different network depths on test data sets
    Structure diagram of improved ResNet18 model
    Schematic diagram of transfer learning method
    Accuracy verification of transfer learning method
    • Table 1. Parameters of real data set

      View table
      View in Article

      Table 1. Parameters of real data set

      Batch No.Actual size/mSize/pixelCutting No.
      11.22×0.1753×7120×850350
      21.22×0.1753×7120×850300
      30.70×0.1753×7120×850300
    • Table 2. Experimental environment and specific parameters

      View table
      View in Article

      Table 2. Experimental environment and specific parameters

      Hardware Environment
      HardwareModel Number
      CPUInter core i7-10750H
      GPUNVIDIA RTX2060
      Memory24 GB
      Software Environment
      SoftwareName
      SystemWindows10
      configurationPytorch 3.6 cuda 10.1
      Training parameters
      ParameterValue
      Batch size8
      Epoch25
      CUDAEnable
    • Table 3. Experimental results of four types of defects by detection system on simulated data sets

      View table
      View in Article

      Table 3. Experimental results of four types of defects by detection system on simulated data sets

      Defect typesPollution defectScratches pollutionBreakage defectLack of design and colorAverage
      Recall/%98.299.210010099.6
      Precision/% 100100100100100
      Accuracy/% 98.899.510010099.6
      Time/ms 305305305305305
    • Table 4. Results comparison of proposed method and other methods on real data sets

      View table
      View in Article

      Table 4. Results comparison of proposed method and other methods on real data sets

      ModelACCLOSS
      ResNet1883.2%0.332
      DenseNet12197.2%0.077
      SqueezeNet95.8%0.097
      MobileNet V392.0%0.118
      Our model98.7%0.011
    Tools

    Get Citation

    Copy Citation Text

    Lixing LIN, Zhenping XIA, Hao XU, Yu SONG, Fuyuan HU. Defect detection on complex texture surface based on optimized ResNet[J]. Journal of Applied Optics, 2023, 44(1): 104

    Download Citation

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

    Category: Research Articles

    Received: Mar. 28, 2022

    Accepted: --

    Published Online: Feb. 22, 2023

    The Author Email: XIA Zhenping (xzp@usts.edu.cn)

    DOI:10.5768/JAO202344.0102006

    Topics