Optical Instruments, Volume. 45, Issue 4, 17(2023)

Design of roughness detection system based on transfer learning and model fusion

Qiang ZHANG... Zhiwen HUANG and Jianmin ZHU* |Show fewer author(s)
Author Affiliations
  • School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • show less
    Figures & Tables(11)
    Roughness detection system
    System inspection page
    Sample roughness image dataset
    Algorithm framework
    Training visualization curve
    FusionNet training curve
    Confusion matrix of identification result
    Comparison of system identification accuracy
    • Table 1. Network fine-tuning parameters

      View table
      View in Article

      Table 1. Network fine-tuning parameters

      模型训练层
      VGGNet-19所有层
      VGGNet-19 Fine-Tuning解冻14~16层
      Inception-V3所有层
      Inception-V3 Fine-Tuning解冻44~46层
      DenseNet121所有层
      DenseNet121 Fine-Tuning解冻119~121层
    • Table 2. Ablation experiment results

      View table
      View in Article

      Table 2. Ablation experiment results

      模型测试集准确率/%
      VGGNet-19 Fine-Tuning79.47
      Inception-V3 Fine-Tuning82.25
      DenseNet121 Fine-Tuning85.43
      FusionNet90.32
    • Table 3. Model comparison results

      View table
      View in Article

      Table 3. Model comparison results

      模型测试集准确率/%
      MobileNetV385.68
      EfficientNetV283.25
      GoogleNet82.36
      Xception81.72
      ResNet5086.21
      FusionNet90.32
    Tools

    Get Citation

    Copy Citation Text

    Qiang ZHANG, Zhiwen HUANG, Jianmin ZHU. Design of roughness detection system based on transfer learning and model fusion[J]. Optical Instruments, 2023, 45(4): 17

    Download Citation

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

    Category: APPLICATION TECHNOLOGY

    Received: Nov. 10, 2022

    Accepted: --

    Published Online: Sep. 26, 2023

    The Author Email: ZHU Jianmin (jmzhu6688@163.com)

    DOI:10.3969/j.issn.1005-5630.2023.004.003

    Topics