• Acta Optica Sinica
  • Vol. 40, Issue 11, 1111002 (2020)
Zhuang Miao1、2, Yong Zhang1、3、*, and Weihua Li1、2
Author Affiliations
  • 1Shanghai Institute of Technical Physics, Chinese Academy of Sciences, Shanghai 200083, China
  • 2University of Chinese Academy of Sciences, Beijing 100049, China
  • 3Key Laboratory of Infrared System Detection and Imaging Technology, Chinese Academy of Sciences, Shanghai 200083, China
  • show less
    DOI: 10.3788/AOS202040.1111002 Cite this Article Set citation alerts
    Zhuang Miao, Yong Zhang, Weihua Li. Infrared Target Modeling Method Based on Double Adversarial Autoencoding Network[J]. Acta Optica Sinica, 2020, 40(11): 1111002 Copy Citation Text show less
    Structure of GAN model
    Fig. 1. Structure of GAN model
    Structure of VAE model
    Fig. 2. Structure of VAE model
    Structure of CDAAE model
    Fig. 3. Structure of CDAAE model
    Network structures of encoder and latent-space discriminator
    Fig. 4. Network structures of encoder and latent-space discriminator
    Network structures of encoder and sample discriminator
    Fig. 5. Network structures of encoder and sample discriminator
    Test results of CDAAE model on infrared dataset. (a) Real samples; (b) reconstructed samples; (c) random samples
    Fig. 6. Test results of CDAAE model on infrared dataset. (a) Real samples; (b) reconstructed samples; (c) random samples
    Test results of CDAAE model on MINIST dataset. (a) Real samples; (b) reconstructed samples; (c) random samples
    Fig. 7. Test results of CDAAE model on MINIST dataset. (a) Real samples; (b) reconstructed samples; (c) random samples
    Hyper parameterInfrared datasetMINIST dataset
    Training loop100000100000
    Class number410
    Batch size64256
    Slope of Leaky ReLU0.010.01
    Learning rate of E/Dg0.000020.00002
    Learning rate of Dl/Ds0.00010.0001
    Data dimension in latent-space84
    Optimization of E/Dg/Dl/Ds per training loop2∶4∶1∶12∶4∶1∶1
    Table 1. Hyper parameters used for CDAAE model training
    ConditionInfrared datasetMINIST dataset
    Group 1 vs group 210.138.85
    Real samples versus reconstructed samples14.267.51
    Real samples versus random samples35.3610.97
    Real samples versus random samples from α-GAN32.4712.18
    Real samples versus random samples from C-DCGAN48.8327.62
    Real samples versus random samples from CVAE52.7425.55
    Table 2. FID evaluation results of CDAAE model
    Number of pictures in Infrared datasetNumber of pictures in MINIST datasetInfrared datasetMINIST dataset
    1200300092.8397.80
    1200+12003000+300094.3398.33
    1200+24003000+600096.0099.26
    Table 3. Accuracy in data augmentation process%
    Zhuang Miao, Yong Zhang, Weihua Li. Infrared Target Modeling Method Based on Double Adversarial Autoencoding Network[J]. Acta Optica Sinica, 2020, 40(11): 1111002
    Download Citation