Optics and Precision Engineering, Volume. 31, Issue 14, 2080(2023)

Generate adversarial network for super-resolution reconstruction of remote sensing images by fusing edge enhancement and non-local modules

Jie LIU1,*... Ruo QI1 and Ke HAN2 |Show fewer author(s)
Author Affiliations
  • 1College of Measurement and Control Technology and Communication Engineering, Harbin University of Science and Technology, Harbin50080, China
  • 2School of Computer and Information Engineering, Harbin University of Commerce, Harbin15008, China
  • show less
    Figures & Tables(15)
    Overall framework of ENGAN algorithm
    Edge-Enhanced Network
    Mask branch before and after improvement
    SR network framework
    Non-local module
    Visual effect comparison
    Comparison of Ledge_cons with edge consistency loss
    Test results
    • Table 1. Comparison of experimental results of different methods

      View table
      View in Article

      Table 1. Comparison of experimental results of different methods

      退化类型指标BicubicESRGANRealSRReal-ESRGANBSRGANENGAN(Ours)
      TypeⅠSSIM0.8140.7650.7580.7900.7660.793
      PSNR27.51126.36225.58725.84725.59126.800
      RMSE12.14114.83314.72514.33514.53812.851
      TypeⅡSSIM0.7590.6820.6900.7360.7040.737
      PSNR25.89923.96724.00524.46424.08626.114
      RMSE14.62418.06217.99417.07117.50415.603
      TypeⅢSSIM0.7550.7010.7240.7480.7320.769
      PSNR26.42724.09324.25125.36125.19426.404
      RMSE13.79917.73917.23215.33815.33513.563
      TypeⅣSSIM0.5470.4730.4950.5690.5570.625
      PSNR23.51222.16422.49623.24023.11024.510
      RMSE19.09820.10819.93319.62719.73817.265
    • Table 2. Algorithm ablation experiments

      View table
      View in Article

      Table 2. Algorithm ablation experiments

      退化类型指标BSRGANEGANEGAN*EGAN**ENGAN
      TypeⅠSSIM0.7660.7740.7840.7840.793
      PSNR25.59126.62526.74226.78326.800
      RMSE14.53813.14712.98612.92712.851
      TypeⅡSSIM0.7040.7200.7280.7300.737
      PSNR24.08625.01725.07925.16326.114
      RMSE17.50415.90115.80015.67815.603
      TypeⅢSSIM0.7320.7510.7580.7610.769
      PSNR25.19426.22526.32726.39226.404
      RMSE15.33513.86413.71513.63313.563
      TypeⅣSSIM0.5570.6200.6220.6230.625
      PSNR23.11024.36324.39724.48224.510
      RMSE19.73817.33217.28517.32217.265
    • Table 3. Comparison of different λ4 results

      View table
      View in Article

      Table 3. Comparison of different λ4 results

      退化类型指标λ4=0.1λ4=0.5λ4=1λ4=2λ4=5λ4=7λ4=10
      TypeⅠSSIM0.7250.7290.7540.7600.7740.7440.732
      PSNR25.48325.78126.28426.44126.62526.09125.774
      RMSE14.69014.55213.64413.41113.14713.98514.298
      TypeⅡSSIM0.6730.6780.7050.7100.7200.6980.679
      PSNR23.81323.90624.91225.00725.01724.40223.920
      RMSE17.97717.82016.05615.90215.90116.10117.782
      TypeⅢSSIM0.7030.7070.7340.7380.7510.7260.709
      PSNR24.81524.84725.89226.01426.22525.61224.856
      RMSE15.91315.86314.34114.15813.86414.68515.839
      TypeⅣSSIM0.5740.5810.6030.6100.6200.5910.585
      PSNR22.87523.13223.92324.14024.36323.50123.294
      RMSE19.54719.16218.14317.74717.33218.75218.980
    • Table 4. Comparison of different λ5 results

      View table
      View in Article

      Table 4. Comparison of different λ5 results

      退化类型指标λ5=0λ5=0.05λ5=0.1λ5=0.5λ5=1λ5=2λ5=5λ5=7λ5=10
      TypeⅣSSIM0.6200.5330.5350.5320.5370.5420.5790.5610.542
      PSNR24.36322.59022.70933.50422.85023.00623.78323.48522.954
      RMSE17.33220.52220.44520.57120.30119.80918.33718.90119.733
    • Table 4. Comparison of different λ5 results

      View table
      View in Article

      Table 4. Comparison of different λ5 results

      退化类型指标λ5=0λ5=0.05λ5=0.1λ5=0.5λ5=1λ5=2λ5=5λ5=7λ5=10
      TypeⅠSSIM0.77407350.7390.7330.7340.7390.7440.7290.703
      PSNR26.62525.18025.38525.37325.38125.68525.94025.39124.752
      RMSE13.14715.37215.52515.44015.41215.37214.0515.50316.002
      TypeⅡSSIM0.7200.6770.6800.6760.6800.6830.6930.6720.650
      PSNR25.01723.62023.74223.60323.81323.92024.60523.56223.098
      RMSE15.90118.36218.14518.07118.00018.14516.52218.51118.840
      TypeⅢSSIM0.7510.7020.7060.7030.7070.7110.7170.6980.677
      PSNR26.22524.50624.76324.58224.80824.82025.53924.49123.706
      RMSE13.86416.17915.96216.09815.94016.92014.80516.99717.959
    • Table 5. Discriminator used for the comparison of image results at different stages

      View table
      View in Article

      Table 5. Discriminator used for the comparison of image results at different stages

      退化类型指标判别IBase判别IHR
      TypeⅠSSIM0.7740.773
      PSNR26.62526.548
      RMSE13.14713.245
      TypeⅡSSIM0.7200.719
      PSNR25.01724.981
      RMSE15.90115.950
      TypeⅢSSIM0.7510.751
      PSNR26.22526.177
      RMSE13.86413.927
      TypeⅣSSIM0.6200.619
      PSNR24.36324.337
      RMSE17.33217.377
    • Table 6. YOLOv5 detection results

      View table
      View in Article

      Table 6. YOLOv5 detection results

      退化

      类型

      LRSR

      mAP

      (IoU=0.5)

      mAP

      (IoU=0.5:0.95)

      mAP

      (IoU=0.5)

      mAP

      (IoU=0.5:0.95)

      TypeⅠ0.6830.3820.6830.383
      TypeⅡ0.6710.3620.6850.370
      TypeⅢ0.5300.2910.6330.351
      TypeⅣ0.1830.08850.3380.188
    Tools

    Get Citation

    Copy Citation Text

    Jie LIU, Ruo QI, Ke HAN. Generate adversarial network for super-resolution reconstruction of remote sensing images by fusing edge enhancement and non-local modules[J]. Optics and Precision Engineering, 2023, 31(14): 2080

    Download Citation

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

    Category: Information Sciences

    Received: Jul. 30, 2022

    Accepted: --

    Published Online: Aug. 2, 2023

    The Author Email: LIU Jie (liujie@hrbust.edu.cn)

    DOI:10.37188/OPE.20233114.2080

    Topics