• Acta Optica Sinica
  • Vol. 41, Issue 12, 1228002 (2021)
Mengyao Wang, Xiangchao Meng*, Feng Shao**, and Randi Fu
Author Affiliations
  • Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, Zhejiang 315211, China
  • show less
    DOI: 10.3788/AOS202141.1228002 Cite this Article Set citation alerts
    Mengyao Wang, Xiangchao Meng, Feng Shao, Randi Fu. SAR-Assisted Optical Remote Sensing Image Cloud Removal Method Based on Deep Learning[J]. Acta Optica Sinica, 2021, 41(12): 1228002 Copy Citation Text show less
    Overall flowchart of proposed method
    Fig. 1. Overall flowchart of proposed method
    Image samples
    Fig. 2. Image samples
    Results of simulation experiments. (a) SAR images; (b) real optical images; (c) simulated optical images contaminated by clouds; (d) experimental results of cGAN model; (e) experimental results of pix2pix model; (f) experimental results of cGAN+SSIM model; (g) experimental results of proposed model
    Fig. 3. Results of simulation experiments. (a) SAR images; (b) real optical images; (c) simulated optical images contaminated by clouds; (d) experimental results of cGAN model; (e) experimental results of pix2pix model; (f) experimental results of cGAN+SSIM model; (g) experimental results of proposed model
    Results of real experiments. (a) SAR images; (b) optical images without cloud cover in a similar time phase; (c) optical remote sensing images with cloud cover; (d) experimental results of cGAN model; (e) experimental results of pix2pix model; (f) experimental results of cGAN+SSIM model; (g) experimental results of proposed model
    Fig. 4. Results of real experiments. (a) SAR images; (b) optical images without cloud cover in a similar time phase; (c) optical remote sensing images with cloud cover; (d) experimental results of cGAN model; (e) experimental results of pix2pix model; (f) experimental results of cGAN+SSIM model; (g) experimental results of proposed model
    DatasetSimulated experimentReal experiment
    SAROpticalSimulatedcorrupted opticalSARSimilar timephase opticalReal corruptedoptical
    Obtain time2020.04.292020.04.282020.04.282020.05.232020.05.132020.06.17
    Table 1. Information about simulated experiments and real experiments
    ModelProposed modelcGAN+SSIMpix2pixcGAN
    SSIMMean0.86550.8604¯0.85580.7241
    95% confidence interval(0.860,0.871)(0.855,0.866)¯(0.850,0.861)(0.718,0.730)
    CCMean0.84430.8399¯0.83490.3732
    95% confidence interval(0.836,0.852)(0.832,0.848)¯(0.827,0.843)(0.359,0.388)
    PSNRMean27.588627.5141¯27.321316.2218
    95% confidence interval(27.218,27.960)(27.126,27.902)¯(26.937,27.705)(16.158,16.285)
    ERGASMean20.224420.2382¯20.832676.6522
    95% confidence interval(19.640,20.809)(19.630,20.846)¯(20.208,21.457)(72.501,80.803)
    SAMMean1.91942.07042.0117¯6.5556
    95% confidence interval(1.874,1.965)(2.028,2.113)(1.979,2.056)¯(6.415,6.696)
    Table 2. Quantitative evaluation results of simulation experiments (mean and 95% confidence interval)
    ModelProposed modelcGAN+SSIMpix2pixcGAN
    SSIMMean0.74670.7371¯0.70730.6149
    95% confidence interval(0.737,0.756)(0.728,0.747)¯(0.697,0.718)(0.601,0.629)
    CCMean0.58200.5655¯0.54550.4398
    95% confidence interval(0.558,0.606)(0.542,0.589)¯(0.522,0.569)(0.415,0.465)
    PSNRMean23.803323.5446¯23.340920.5477
    95% confidence interval(23.448,24.158)(23.199,23.891)¯(22.998,23.684)(20.215,20.880)
    ERGASMean29.971931.0080¯31.420243.5042
    95% confidence interval(28.981,30.963)(30.062,31.955)¯(30.453,32.388)(42.023,44.986)
    SAMMean4.34234.70614.5815¯5.3864
    95% confidence interval(4.202,4.483)(4.570,4.842)(4.440,4.723)¯(5.213,5.560)
    Table 3. Quantitative evaluation results of real experiments (mean and 95% confidence interval)
    Mengyao Wang, Xiangchao Meng, Feng Shao, Randi Fu. SAR-Assisted Optical Remote Sensing Image Cloud Removal Method Based on Deep Learning[J]. Acta Optica Sinica, 2021, 41(12): 1228002
    Download Citation