Acta Photonica Sinica, Volume. 53, Issue 11, 1112001(2024)

Speckle Image Deformation Measurement Method Based on Convolutional Neural Network UNet++

Qiang CHEN... Junzhe LIANG and Jin LIANG* |Show fewer author(s)
Author Affiliations
  • State Key Laboratory of Precision Manufacturing Technology,School of Mechanical Engineering,Xi'an Jiaotong University,Xi'an 710049,China
  • show less
    Figures & Tables(14)
    Four kinds of speckle images
    Image pairs and displacement fields at different spatial frequencies
    Schematic diagram of the structure of DIC-Net++
    Schematic diagram of the structure of residual block
    Schematic diagram of the structure of CA Block
    Performance of different models on self-built dataset
    Displacement and noise level on Star5 predicted by three networks
    Spatial resolution on Star5 predicted by three networks
    Performance of different models on Star6
    • Table 1. The training results of DIC-Net++

      View table
      View in Article

      Table 1. The training results of DIC-Net++

      AEE/pixelDIC-Net++-dDIC-Net++-s
      Train set0.033 20.009 8
      Validation set0.033 90.010 3
    • Table 2. Performance of DIC-Net++-d on self-built dataset

      View table
      View in Article

      Table 2. Performance of DIC-Net++-d on self-built dataset

      GroupModelRMSE/pixelMAE/pixel
      uvuv
      0StrainNet-f0.036 540.035 850.027 660.026 89
      DIC-Net-d0.025 230.025 660.019 250.019 28
      DIC-Net++-d0.015 480.015 070.011 260.011 20
      1StrainNet-f0.118 560.099 650.070 210.079 55
      DIC-Net-d0.083 030.085 640.050 140.064 04
      DIC-Net++-d0.016 170.016 000.012 380.012 19
    • Table 3. Performance of different models on DIC Challenge 1.0

      View table
      View in Article

      Table 3. Performance of different models on DIC Challenge 1.0

      ModelRMSE/pixel
      Sample 1Sample 3Sample 4
      StrainNet-f0.150 500.053 520.148 14
      DIC-Net-d0.236 440.020 890.106 48
      DIC-Net++-d0.067 840.018 280.085 58
    • Table 4. Performance of different models on Star5

      View table
      View in Article

      Table 4. Performance of different models on Star5

      ModelSRσvMEI
      SrainNet-f2850.022 66.441
      DIC-Net-d2310.015 33.534
      DIC-Net++-d980.014 01.372
    • Table 5. Performance of different models on Star6

      View table
      View in Article

      Table 5. Performance of different models on Star6

      ModelSRσvMEI
      DIC-Net-s820.009 9360.008 1
      DIC-Net++-s1240.005 459 20.003 7
    Tools

    Get Citation

    Copy Citation Text

    Qiang CHEN, Junzhe LIANG, Jin LIANG. Speckle Image Deformation Measurement Method Based on Convolutional Neural Network UNet++[J]. Acta Photonica Sinica, 2024, 53(11): 1112001

    Download Citation

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

    Category: Instrumentation, Measurement and Metrology

    Received: Apr. 12, 2024

    Accepted: May. 24, 2024

    Published Online: Jan. 8, 2025

    The Author Email: LIANG Jin (liangjin@mail.edu.cn)

    DOI:10.3788/gzxb20245311.1112001

    Topics