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
Figures & Tables(15)
Table 1. Comparison of experimental results of different methods
View tableView in ArticleTable 1. Comparison of experimental results of different methods
退化类型 指标 Bicubic ESRGAN RealSR Real-ESRGAN BSRGAN ENGAN（Ours） TypeⅠ SSIM 0.814 0.765 0.758 0.790 0.766 0.793 PSNR 27.511 26.362 25.587 25.847 25.591 26.800 RMSE 12.141 14.833 14.725 14.335 14.538 12.851 TypeⅡ SSIM 0.759 0.682 0.690 0.736 0.704 0.737 PSNR 25.899 23.967 24.005 24.464 24.086 26.114 RMSE 14.624 18.062 17.994 17.071 17.504 15.603 TypeⅢ SSIM 0.755 0.701 0.724 0.748 0.732 0.769 PSNR 26.427 24.093 24.251 25.361 25.194 26.404 RMSE 13.799 17.739 17.232 15.338 15.335 13.563 TypeⅣ SSIM 0.547 0.473 0.495 0.569 0.557 0.625 PSNR 23.512 22.164 22.496 23.240 23.110 24.510 RMSE 19.098 20.108 19.933 19.627 19.738 17.265
Table 2. Algorithm ablation experiments
View tableView in ArticleTable 2. Algorithm ablation experiments
退化类型 指标 BSRGAN EGAN EGAN* EGAN** ENGAN TypeⅠ SSIM 0.766 0.774 0.784 0.784 0.793 PSNR 25.591 26.625 26.742 26.783 26.800 RMSE 14.538 13.147 12.986 12.927 12.851 TypeⅡ SSIM 0.704 0.720 0.728 0.730 0.737 PSNR 24.086 25.017 25.079 25.163 26.114 RMSE 17.504 15.901 15.800 15.678 15.603 TypeⅢ SSIM 0.732 0.751 0.758 0.761 0.769 PSNR 25.194 26.225 26.327 26.392 26.404 RMSE 15.335 13.864 13.715 13.633 13.563 TypeⅣ SSIM 0.557 0.620 0.622 0.623 0.625 PSNR 23.110 24.363 24.397 24.482 24.510 RMSE 19.738 17.332 17.285 17.322 17.265
Table 3. Comparison of different
λ 4 resultsView tableView in ArticleTable 3. Comparison of different
λ 4 results退化类型 指标 λ4=0.1 λ4=0.5 λ4=1 λ4=2 λ4=5 λ4=7 λ4=10 TypeⅠ SSIM 0.725 0.729 0.754 0.760 0.774 0.744 0.732 PSNR 25.483 25.781 26.284 26.441 26.625 26.091 25.774 RMSE 14.690 14.552 13.644 13.411 13.147 13.985 14.298 TypeⅡ SSIM 0.673 0.678 0.705 0.710 0.720 0.698 0.679 PSNR 23.813 23.906 24.912 25.007 25.017 24.402 23.920 RMSE 17.977 17.820 16.056 15.902 15.901 16.101 17.782 TypeⅢ SSIM 0.703 0.707 0.734 0.738 0.751 0.726 0.709 PSNR 24.815 24.847 25.892 26.014 26.225 25.612 24.856 RMSE 15.913 15.863 14.341 14.158 13.864 14.685 15.839 TypeⅣ SSIM 0.574 0.581 0.603 0.610 0.620 0.591 0.585 PSNR 22.875 23.132 23.923 24.140 24.363 23.501 23.294 RMSE 19.547 19.162 18.143 17.747 17.332 18.752 18.980
Table 4. Comparison of different
λ 5 resultsView tableView in ArticleTable 4. Comparison of different
λ 5 results退化类型 指标 λ5=0 λ5=0.05 λ5=0.1 λ5=0.5 λ5=1 λ5=2 λ5=5 λ5=7 λ5=10 TypeⅣ SSIM 0.620 0.533 0.535 0.532 0.537 0.542 0.579 0.561 0.542 PSNR 24.363 22.590 22.709 33.504 22.850 23.006 23.783 23.485 22.954 RMSE 17.332 20.522 20.445 20.571 20.301 19.809 18.337 18.901 19.733
Table 4. Comparison of different
λ 5 resultsView tableView in ArticleTable 4. Comparison of different
λ 5 results退化类型 指标 λ5=0 λ5=0.05 λ5=0.1 λ5=0.5 λ5=1 λ5=2 λ5=5 λ5=7 λ5=10 TypeⅠ SSIM 0.774 0735 0.739 0.733 0.734 0.739 0.744 0.729 0.703 PSNR 26.625 25.180 25.385 25.373 25.381 25.685 25.940 25.391 24.752 RMSE 13.147 15.372 15.525 15.440 15.412 15.372 14.05 15.503 16.002 TypeⅡ SSIM 0.720 0.677 0.680 0.676 0.680 0.683 0.693 0.672 0.650 PSNR 25.017 23.620 23.742 23.603 23.813 23.920 24.605 23.562 23.098 RMSE 15.901 18.362 18.145 18.071 18.000 18.145 16.522 18.511 18.840 TypeⅢ SSIM 0.751 0.702 0.706 0.703 0.707 0.711 0.717 0.698 0.677 PSNR 26.225 24.506 24.763 24.582 24.808 24.820 25.539 24.491 23.706 RMSE 13.864 16.179 15.962 16.098 15.940 16.920 14.805 16.997 17.959
Table 5. Discriminator used for the comparison of image results at different stages
View tableView in ArticleTable 5. Discriminator used for the comparison of image results at different stages
退化类型 指标 判别IBase 判别IHR TypeⅠ SSIM 0.774 0.773 PSNR 26.625 26.548 RMSE 13.147 13.245 TypeⅡ SSIM 0.720 0.719 PSNR 25.017 24.981 RMSE 15.901 15.950 TypeⅢ SSIM 0.751 0.751 PSNR 26.225 26.177 RMSE 13.864 13.927 TypeⅣ SSIM 0.620 0.619 PSNR 24.363 24.337 RMSE 17.332 17.377
Table 6. YOLOv5 detection results
View tableView in ArticleTable 6. YOLOv5 detection results
退化
类型
LR SR mAP
（IoU=0.5）
mAP
（IoU=0.5：0.95）
mAP
（IoU=0.5）
mAP
（IoU=0.5：0.95）
TypeⅠ 0.683 0.382 0.683 0.383 TypeⅡ 0.671 0.362 0.685 0.370 TypeⅢ 0.530 0.291 0.633 0.351 TypeⅣ 0.183 0.0885 0.338 0.188
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
Paper Information
Category: Information Sciences
Received: Jul. 30, 2022
Accepted: --
Published Online: Aug. 2, 2023
The Author Email: LIU Jie (liujie@hrbust.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20