Optics and Precision Engineering, Volume. 31, Issue 16, 2406(2023)

Equipment fault dataset amplification method combine 3D model with improved CycleGAN

Baoping LI... Hengyi QI*, Manli WANG and Po WEI |Show fewer author(s)
Author Affiliations
  • College of Physics and Electronic Information, Henan Polytechnic University, Jiaozuo454000, China
  • show less
    Figures & Tables(15)
    Framework of CycleGAN
    Framework of improved generator
    Feature map of improved generator
    Several generator simplification architectures
    Contour feature images
    Framework of discriminator
    Loss function
    Framework of 3D modeling
    Camera perspective
    Fake Y for multiple network outputs
    MS_SSIM metrics for three generators
    Loss curve for three generators
    • Table 1. Result of three evaluation indicators

      View table
      View in Article

      Table 1. Result of three evaluation indicators

      Generator/GANIS ⬆FID ⬇PSNR ⬆
      DCGAN1.385 7178.805 5/
      WGAN1.375 3161.486 5/
      WGAN-GP1.417 4155.687 1/
      StyleGAN1.397 6164.547 3/
      ChipGAN1.364 1287.971 226.467 4
      ResNet1.484 9232.266 934.595 4
      U-net1.446 9227.299 133.245 7
      Ours1.519 7192.021 235.096 4
    • Table 2. Results of Yolov5 detection

      View table
      View in Article

      Table 2. Results of Yolov5 detection

      数据集左偏AP右偏APMAP
      真实86.795.493.6
      WGAN-GP+真实89.896.194.8
      U-net+真实91.898.695.7
      ResNet+真实93.399.196.2
      Ours+真实96.799.598.1
    • Table 3. Generator parameters and training memory occupancy

      View table
      View in Article

      Table 3. Generator parameters and training memory occupancy

      ModelParams size /MBLayers of GGPU Memory Usage/GB
      ResNet43.42925.0
      U-net207.57544.6
      U-ResNet41.7755.8
    Tools

    Get Citation

    Copy Citation Text

    Baoping LI, Hengyi QI, Manli WANG, Po WEI. Equipment fault dataset amplification method combine 3D model with improved CycleGAN[J]. Optics and Precision Engineering, 2023, 31(16): 2406

    Download Citation

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

    Category: Information Sciences

    Received: Nov. 16, 2022

    Accepted: --

    Published Online: Sep. 5, 2023

    The Author Email: QI Hengyi (mystericq@home.hpu.edu.cn)

    DOI:10.37188/OPE.20233116.2406

    Topics