Optics and Precision Engineering, Volume. 31, Issue 22, 3357(2023)

A lightweight deep learning model for TFT-LCD circuits defect classification based on swin transformer

Yan XIA1... Chen LUO1,*, Yijun ZHOU1 and Lei JIA2 |Show fewer author(s)
Author Affiliations
  • 1School of Mechanical Engineering, Southeast University, Nanjing289, China
  • 2Wuxi Shangshi-finevision Technology Co., Ltd, Wuxi14174, China
  • show less
    Figures & Tables(15)
    Architecture of Swin-T model
    Architecture of Swin Transformer Block
    Token merging module
    Token merging for visualization
    Conventional convolution and depthwise separable convolution
    Knowledge distillation flow
    Loss curve
    Different effect of the parameters T on the model accuracy
    Different effect of the parameters α on the model accuracy at T=3
    Comparison of detection performance between ResNet-34 model and the improved model
    Detection results of the modified model
    • Table 1. Results of parameter n affects experiments

      View table
      View in Article

      Table 1. Results of parameter n affects experiments

      nTop-1 AccFLOPs/Gi/(m·s-1
      094.017.551.42
      192.116.354.49
      290.614.761.92
    • Table 2. Results of ablation experiments

      View table
      View in Article

      Table 2. Results of ablation experiments

      ModelTop-1 Acc/%PrecisionRecallF1FLOPs/Gi/(m·s-1
      Swin-T94.00.935 40.935 60.935 517.551.42
      +Token merging90.60.900 20.901 80.901 014.761.92
      +DW91.20.913 80.891 30.902 414.960.43
      +KD92.70.916 50.905 60.911 014.960.43
    • Table 3. Results of comparison experiments

      View table
      View in Article

      Table 3. Results of comparison experiments

      ModelTop-1 Acc/%FLOPs/Gi/(m·s-1
      Resnet-3487.414.668.97
      DenseNet16991.113.732.72
      EffecientNet-v290.611.628.57
      MobileViT-v290.59.873.95
      ConvNext-T95.117.850.36
      EVA0295.887.510.62
      Ours92.714.960.43
    • Table 4. Results of comparison experiments on public dataset

      View table
      View in Article

      Table 4. Results of comparison experiments on public dataset

      ModelSizeTop-1 Acc/%Params/MFLOPs/G
      EVA-G/1433689.51013.01445.56
      ViT-H/1433688.6632.46363.64
      ConvNext-XL38487.7350.20179.03
      Swin-L38487.1196.74100.28
      ResNet-101d32083.044.5723.82
      MobileViT-v238482.914.2512.35
      EffecientNet-v228882.213.658.16
      Ours38483.327.5013.89
    Tools

    Get Citation

    Copy Citation Text

    Yan XIA, Chen LUO, Yijun ZHOU, Lei JIA. A lightweight deep learning model for TFT-LCD circuits defect classification based on swin transformer[J]. Optics and Precision Engineering, 2023, 31(22): 3357

    Download Citation

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

    Category:

    Received: Mar. 30, 2023

    Accepted: --

    Published Online: Dec. 29, 2023

    The Author Email: LUO Chen (chenluo@seu.edu.cn)

    DOI:10.37188/OPE.20233122.3357

    Topics