Acta Photonica Sinica, Volume. 51, Issue 12, 1210004(2022)

Infrared Ship Image Generation Algorithm Based on ISE-StyleGAN

Haijun LI1... Fancheng KONG1,*, Junjie MU1, Xiao LIU2, Zhenbin DU2 and Yun LIN3 |Show fewer author(s)
Author Affiliations
  • 1Coastal Defense College,Naval Aviation University,Yantai,Shandong 264001,China
  • 2School of Computing,Yantai University,Yantai,Shandong 264005,China
  • 3Office of Academic Affairs,Yantai University,Yantai,Shandong 264005,China
  • show less
    Figures & Tables(14)
    Generative adversarial network structure diagram
    StyleGAN generator structure
    Self-attention
    Improved generator architecture
    Structure of wavelet discriminator
    The training process of ISE-StyleGAN
    Compare of the original images with the generated images
    Comparison of images generated by different generative adversarial network algorithms
    mAP value of each dataset by different algorithms
    • Table 1. Experimental training parameters

      View table
      View in Article

      Table 1. Experimental training parameters

      ParameterValue
      Learning rateTTUR
      OptimizerAdam
      Base size64
      Epoch900
    • Table 2. PSNR index results of generative adversarial network model

      View table
      View in Article

      Table 2. PSNR index results of generative adversarial network model

      AlgorithmPSNR/dB
      Single scaleMultiscaleSmall targetCloud and mist interferenceNight target
      DCGAN13.2547.3636.78410.4369.473
      CycleGAN9.7928.1876.3498.3429.754
      StyleGAN8.8357.8294.3824.9938.438
      ISE-StyleGAN21.85319.34113.74011.28415.584
    • Table 3. MS-SSIM index results of generative adversarial network model

      View table
      View in Article

      Table 3. MS-SSIM index results of generative adversarial network model

      AlgorithmMS-SSIM/ %
      Single scaleMultiscaleSmall targetCloud and mist interferenceNight target
      DCGAN91.685.783.490.293.6
      CycleGAN83.888.184.779.089.2
      StyleGAN82.387.281.973.386.4
      ISE-StyleGAN97.994.589.392.798.5
    • Table 4. Data set composition of object detection test

      View table
      View in Article

      Table 4. Data set composition of object detection test

      NumberAlgorithmSource dataConventionally generated dataAlgorithmic data generationTotal
      1None1 000001 000
      2Conventional expansion1 0002 00003 000
      3DCGAN1 0001 0001 0003 000
      4CycleGAN1 0001 0001 0003 000
      5StyleGAN1 0001 0001 0003 000
      6ISE-StyleGAN1 0001 0001 0003 000
    • Table 5. mAP value of target detection in each dataset

      View table
      View in Article

      Table 5. mAP value of target detection in each dataset

      NumberAlgorithmYOLOv3SSDFaster R⁃CNNCenternet
      1None0.6790.6410.7030.727
      2Conventional expansion0.7710.7330.7880.791
      3DCGAN0.7840.7120.8040.786
      4CycleGAN0.7880.7450.8010.787
      5StyleGAN0.7620.7350.7970.790
      6ISE-StyleGAN0.8330.8120.8220.817
    Tools

    Get Citation

    Copy Citation Text

    Haijun LI, Fancheng KONG, Junjie MU, Xiao LIU, Zhenbin DU, Yun LIN. Infrared Ship Image Generation Algorithm Based on ISE-StyleGAN[J]. Acta Photonica Sinica, 2022, 51(12): 1210004

    Download Citation

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

    Category:

    Received: May. 18, 2022

    Accepted: Jul. 6, 2022

    Published Online: Feb. 6, 2023

    The Author Email: KONG Fancheng (879445564@qq.com)

    DOI:10.3788/gzxb20225112.1210004

    Topics