• Chinese Optics Letters
  • Vol. 19, Issue 8, 081101 (2021)
Zhenming Yu, Zhenyu Ju, Xinlei Zhang, Ziyi Meng, Feifei Yin, and Kun Xu*
Author Affiliations
  • State Key Laboratory of Information Photonics and Optical Communications, Beijing University of Posts and Telecommunications, Beijing 100876, China
  • show less
    DOI: 10.3788/COL202119.081101 Cite this Article Set citation alerts
    Zhenming Yu, Zhenyu Ju, Xinlei Zhang, Ziyi Meng, Feifei Yin, Kun Xu. High-speed multimode fiber imaging system based on conditional generative adversarial network[J]. Chinese Optics Letters, 2021, 19(8): 081101 Copy Citation Text show less
    Structure of the conditional GAN; (a) architecture of the generator; (b) principle of the discriminator. G, generator; D, discriminator.
    Fig. 1. Structure of the conditional GAN; (a) architecture of the generator; (b) principle of the discriminator. G, generator; D, discriminator.
    Experiment setup. DMD, digital micromirror device; OBJ, microscope objective lens; MMF, multimode fiber; CCD, charge-coupled device.
    Fig. 2. Experiment setup. DMD, digital micromirror device; OBJ, microscope objective lens; MMF, multimode fiber; CCD, charge-coupled device.
    Structures of generator in the (a) MNIST experiment and (b) Fashion-MNIST experiment.
    Fig. 3. Structures of generator in the (a) MNIST experiment and (b) Fashion-MNIST experiment.
    Structures of discriminator in the (a) MNIST experiment and (b) Fashion-MNIST experiment.
    Fig. 4. Structures of discriminator in the (a) MNIST experiment and (b) Fashion-MNIST experiment.
    Reconstruction performances with different output resolutions of the discriminator in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    Fig. 5. Reconstruction performances with different output resolutions of the discriminator in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    Loss for training process of U-net and the conditional GAN in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    Fig. 6. Loss for training process of U-net and the conditional GAN in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    Reconstruction results of U-Net and the conditional GAN in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    Fig. 7. Reconstruction results of U-Net and the conditional GAN in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    PSNR and SSIM at each training set number by U-net and the conditional GAN in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    Fig. 8. PSNR and SSIM at each training set number by U-net and the conditional GAN in (a) the MNIST experiment and (b) the Fashion-MNIST experiment.
    Number of Convolutional LayersKernel SizeStrideOutput ResolutionReceptive Field
    11×1132×321×1
    14×4216×164×4
    24×428×810×10
    34×424×422×22
    44×422×246×46
    54×421×194×94
    Table 1. Relationship between the Receptive Field and Convolutional Layer
    Zhenming Yu, Zhenyu Ju, Xinlei Zhang, Ziyi Meng, Feifei Yin, Kun Xu. High-speed multimode fiber imaging system based on conditional generative adversarial network[J]. Chinese Optics Letters, 2021, 19(8): 081101
    Download Citation