Optics and Precision Engineering, Volume. 33, Issue 8, 1238(2025)
Spatial adaptation and frequency fusion network for single remote sensing image super-resolution
Figures & Tables(17)
Fig. 3. （a） Global Multi-scale Field Selection Block； （b） Channel Non-local Attention
Fig. 4. （a） Spatial adaptively selection block； （b） Frequency separation selection block； （c） Spatial non-local attention
Fig. 5. Visualization of learned feature maps from the HSFM at different scales
Fig. 7. Super-resolved outputs of different methods on UCMerced dataset
Fig. 8. Super-resolved outputs of different methods on AID dataset
Fig. 9. Super-resolved outputs of different methods on Sentinel-2 data
Fig. 10. Super-resolved outputs of different methods on HSRS-SC dataset
Table 1. Ablation experimental results of SAF2Net on UCMerced dataset with×4 scale factor
View tableView in ArticleTable 1. Ablation experimental results of SAF2Net on UCMerced dataset with×4 scale factor
Ablation Variant PSNR SSIM SAF2Net - 28.13 0.7748 GHFEM （1） HSFM： w/o scale8 27.98 0.7716 （2） HSFM： w/o scale8+scale4 27.97 0.7709 （3） HSFM： w/o scale8 + scale4+scale2 27.98 0.7707 （4） HSFM： w/o CMG 27.89 0.7678 （5） GMFS： w/o ASM 28.01 0.7702 LEFEM （6） w/o SASB 27.91 0.7690 （7） w/o FSSB 27.99 0.7710 （8） w/o SASB + FSSB 27.86 0.7669 （9） w/o CMG 28.04 0.7730 Loss （10） w/o LFreq 27.96 0.7709
Table 2. Evaluation metrics of different methods over the UCMerced test dataset
View tableView in ArticleTable 2. Evaluation metrics of different methods over the UCMerced test dataset
Scale Metric Bicubic SRCNN VDSR LGCNet DCM TransENet HSENet HAUNet TTST TSFNet 本文 UCMerced
×2
PSNR 30.76 32.84 33.87 33.48 33.65 34.03 34.22 34.46 34.39 34.48 34.53 SSIM 0.878 9 0.915 2 0.927 4 0.923 5 0.927 4 0.930 1 0.932 7 0.933 3 0.932 4 0.933 5 0.934 2 SCC 0.507 9 0.595 6 0.619 6 0.605 6 0.631 8 0.636 6 0.634 1 0.643 7 0.645 5 0.646 1 0.647 8 SAM 0.071 2 0.055 6 0.051 9 0.054 0 0.050 4 0.049 8 0.050 0 0.048 8 0.048 5 0.048 4 0.047 6 UCMerced
×3
PSNR 27.46 28.66 29.75 29.28 29.52 29.92 30.00 30.34 30.25 30.32 30.45 SSIM 0.763 1 0.803 8 0.834 6 0.823 8 0.839 4 0.840 8 0.842 0 0.847 6 0.845 8 0.847 2 0.850 9 SCC 0.276 6 0.358 1 0.394 1 0.365 4 0.405 5 0.407 6 0.413 1 0.423 6 0.420 1 0.424 8 0.437 2 SAM 0.103 9 0.088 3 0.082 9 0.086 8 0.082 0 0.080 7 0.080 6 0.077 9 0.078 7 0.078 1 0.076 6 UCMerced
×4
PSNR 25.65 26.78 27.54 27.02 27.22 27.77 27.73 28.06 27.98 27.92 28.13 SSIM 0.672 5 0.721 9 0.752 2 0.733 3 0.752 8 0.763 0 0.762 3 0.772 6 0.769 8 0.769 5 0.774 8 SCC 0.150 0 0.222 5 0.258 9 0.224 6 0.261 2 0.270 7 0.269 2 0.293 2 0.289 2 0.292 0 0.301 5 SAM 0.126 9 0.111 8 0.105 5 0.110 6 0.105 3 0.102 7 0.104 3 0.099 7 0.100 6 0.100 7 0.098 1
Table 3. Evaluation metrics of different methods over the AID test dataset
View tableView in ArticleTable 3. Evaluation metrics of different methods over the AID test dataset
Scale Metric Bicubic SRCNN VDSR LGCNet DCM TransENet HSENet HAUNet TTST TSFNet 本文 AID
×2
PSNR 32.39 34.49 35.11 34.80 35.21 35.28 35.50 35.58 35.60 35.54 35.67 SSIM 0.890 6 0.928 6 0.934 0 0.932 0 0.936 6 0.937 4 0.938 3 0.938 9 0.939 2 0.938 5 0.940 3 SCC 0.520 2 0.600 6 0.618 1 0.611 6 0.639 0 0.652 9 0.662 6 0.661 6 0.663 8 0.659 8 0.666 3 SAM 0.074 2 0.058 0 0.054 4 0.055 8 0.053 2 0.052 9 0.052 4 0.051 8 0.051 3 0.052 0 0.050 1 AID
×3
PSNR 29.08 30.55 31.17 30.73 31.31 31.45 31.49 31.64 31.62 31.54 31.76 SSIM 0.786 3 0.837 2 0.851 1 0.841 7 0.856 1 0.859 5 0.858 8 0.861 9 0.862 6 0.859 5 0.863 5 SCC 0.274 3 0.353 4 0.380 0 0.356 4 0.400 5 0.409 2 0.405 3 0.414 4 0.416 0 0.409 6 0.418 9 SAM 0.105 5 0.089 4 0.083 6 0.087 6 0.082 0 0.080 5 0.080 6 0.079 2 0.079 0 0.080 3 0.078 0 AID
×4
PSNR 27.30 28.40 28.99 28.61 29.17 29.38 29.32 29.49 29.40 29.50 29.59 SSIM 0.703 6 0.756 1 0.775 3 0.762 6 0.782 4 0.790 9 0.786 7 0.792 4 0.789 0 0.792 6 0.794 5 SCC 0.141 6 0.211 1 0.242 7 0.218 2 0.267 5 0.284 3 0.276 5 0.293 5 0.292 2 0.296 0 0.298 3 SAM 0.127 5 0.112 4 0.105 5 0.109 9 0.103 2 0.100 5 0.101 7 0.099 7 0.100 3 0.099 6 0.098 7
Table 4. PSNR/SSIM of different methods on 21 types of RS scenes on UCMerced test dataset for×3 SR
View tableView in ArticleTable 4. PSNR/SSIM of different methods on 21 types of RS scenes on UCMerced test dataset for×3 SR
Class Baseline 自然图像超分方法 Bicubic SRCNN VDSR PSNR SSIM PSNR SSIM PSNR SSIM Agricultural 26.86 0.542 4 27.47 0.591 4 27.62 0.611 1 Airplane 26.71 0.783 7 28.24 0.833 2 29.91 0.845 9 Baseball diamond 33.33 0.835 1 34.33 0.867 1 35.09 0.881 6 Beach 36.14 0.896 7 37.00 0.915 5 37.43 0.918 8 Buildings 25.09 0.774 3 26.84 0.827 6 28.57 0.860 6 Chaparal 25.21 0.700 2 26.11 0.745 3 26.78 0.775 2 D-Residential 25.76 0.773 3 27.41 0.818 4 28.84 0.860 2 Forest 27.53 0.703 5 28.24 0.740 4 28.56 0.766 5 FreeWay 27.36 0.766 7 28.69 0.803 3 30.33 0.838 7 Golf Course 35.21 0.881 5 36.15 0.890 3 36.40 0.901 7
Table 5. PSNR/SSIM of different methods on 30 types of RS scenes on AID test dataset for ×4 SR
View tableView in ArticleTable 5. PSNR/SSIM of different methods on 30 types of RS scenes on AID test dataset for ×4 SR
Baseline 自然图像超分方法 Class Bicubic SRCNN VDSR PSNR SSIM PSNR SSIM PSNR SSIM Airport 27.03 0.734 4 28.17 0.781 3 28.82 0.801 2 Bare Land 34.88 0.827 8 35.63 0.853 5 35.98 0.858 3 Baseball Field 29.06 0.787 8 30.51 0.827 5 31.18 0.840 3 Beach 31.07 0.785 0 31.92 0.813 6 32.29 0.819 5 Bridge 28.98 0.791 5 30.41 0.828 2 31.19 0.842 1 Center 25.26 0.668 1 26.59 0.729 8 27.48 0.760 1 Church 22.15 0.558 6 23.41 0.644 8 24.12 0.679 4 Commercial 25.83 0.693 5 27.05 0.755 8 27.62 0.777 2 D-Residential 23.05 0.591 2 24.13 0.669 1 24.70 0.699 3 Desert 38.49 0.900 9 38.84 0.913 0 39.13 0.914 8 Farmland 32.30 0.808 1 33.48 0.839 3 34.20 0.852 9 Forest 27.39 0.619 9 28.15 0.681 7 28.36 0.690 2 Industrial 24.75 0.655 2 26.00 0.721 8 26.72 0.751 5 Meadow 32.06 0.707 0 32.57 0.735 3 32.77 0.739 5 M-Residential 26.09 0.643 6 27.37 0.711 3 28.06 0.735 8 Mountain 28.04 0.689 0 28.90 0.739 8 29.11 0.747 7 Park 26.23 0.681 2 27.25 0.738 3 27.69 0.754 9 Parking 22.33 0.671 5 24.01 0.754 9 25.21 0.795 6 Playground 27.27 0.734 8 28.72 0.784 6 29.62 0.809 8 Pond 28.94 0.771 5 29.85 0.804 9 30.26 0.814 6 Port 24.69 0.742 2 25.82 0.790 7 26.43 0.809 5 Railway Station 26.31 0.668 6 27.55 0.732 4 28.19 0.757 8 Resort 25.98 0.694 2 27.12 0.748 4 27.71 0.768 1 River 29.61 0.725 1 30.48 0.765 2 30.82 0.775 0 School 24.91 0.667 3 26.13 0.731 1 26.78 0.755 5 S-Residential 25.41 0.553 6 26.16 0.616 5 26.46 0.630 0 Square 26.75 0.705 8 28.13 0.764 0 28.91 0.787 5 Stadium 24.81 0.689 3 26.10 0.746 7 26.88 0.774 0 Storage Tanks 24.18 0.636 1 25.27 0.695 9 25.86 0.721 7 Viaduct 25.86 0.649 0 27.03 0.712 3 27.74 0.741 8 Average 27.30 0.703 6 28.40 0.756 1 28.99 0.775 3
Table 5. PSNR/SSIM of different methods on 21 types of RS scenes on UCMerced test dataset for×3 SR
View tableView in ArticleTable 5. PSNR/SSIM of different methods on 21 types of RS scenes on UCMerced test dataset for×3 SR
Class Baseline 自然图像超分方法 Bicubic SRCNN VDSR PSNR SSIM PSNR SSIM PSNR SSIM Harbor 21.25 0.754 8 22.82 0.822 8 24.39 0.889 2 Intersection 26.48 0.760 4 27.67 0.818 4 28.82 0.850 7 M-Residential 25.68 0.746 6 27.06 0.807 4 28.43 0.845 3 Mobile Homepark 22.25 0.693 9 23.89 0.745 9 25.25 0.804 4 Overpass 24.59 0.706 4 25.65 0.735 6 27.78 0.802 0 Parking Lot 21.75 0.742 5 23.11 0.788 9 24.40 0.829 6 River 28.12 0.720 4 28.89 0.745 1 29.18 0.791 9 Runway 29.30 0.764 2 30.61 0.808 9 31.48 0.827 1 S-Residential 28.34 0.802 8 29.40 0.834 3 31.44 0.861 8 Storage Tanks 29.97 0.835 2 31.33 0.870 2 32.77 0.894 6 Tennis Court 29.75 0.839 8 30.98 0.867 2 32.36 0.888 2 Average 27.46 0.763 1 28.66 0.803 8 29.75 0.834 6 Class 遥感图像超分方法 TransENet HSENet HAUNet TTST TSFNet 本文 PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM Agricultural 28.02 0.622 8 27.64 0.613 9 28.07 0.626 0 27.67 0.613 3 27.90 0.615 4 28.04 0.634 6 Airplane 29.94 0.853 4 30.09 0.852 8 30.41 0.856 2 30.45 0.857 8 30.50 0.858 9 30.33 0.859 2 Baseball diamond 35.04 0.882 8 35.05 0.882 8 35.34 0.884 1 35.35 0.885 2 35.37 0.885 5 35.39 0.885 7 Beach 37.53 0.920 4 37.69 0.921 4 37.90 0.923 7 37.69 0.922 1 37.73 0.922 2 37.88 0.924 1 Buildings 28.81 0.869 1 28.95 0.869 5 29.47 0.876 8 29.23 0.871 9 29.45 0.874 9 29.55 0.878 5 Chaparal 26.69 0.778 2 26.70 0.780 0 26.86 0.785 4 26.90 0.784 6 26.93 0.785 9 26.98 0.786 6 D-Residential 29.11 0.869 2 29.24 0.869 2 29.63 0.875 1 29.33 0.870 5 29.55 0.874 1 29.77 0.878 7 Forest 28.59 0.770 2 28.59 0.771 0 28.69 0.775 7 28.65 0.775 2 28.70 0.775 1 28.73 0.776 6 FreeWay 30.38 0.843 6 30.63 0.846 3 30.96 0.855 1 30.90 0.852 0 30.98 0.854 9 31.16 0.859 3 Golf Course 36.68 0.903 8 36.62 0.903 8 36.91 0.905 0 36.75 0.904 5 36.78 0.904 7 36.87 0.905 9 Harbor 24.72 0.900 8 24.88 0.904 8 25.51 0.916 3 25.34 0.910 7 25.46 0.913 5 25.71 0.919 3 Intersection 29.03 0.857 3 29.21 0.859 0 29.52 0.865 0 29.42 0.865 2 29.50 0.865 9 29.71 0.871 0 M-Residential 28.47 0.846 8 28.55 0.849 7 28.91 0.855 9 28.95 0.857 2 29.12 0.857 4 29.01 0.858 6 Mobile Homepark 25.64 0.816 1 25.70 0.818 2 26.14 0.826 0 26.06 0.823 5 26.25 0.827 6 26.31 0.831 7 Overpass 27.83 0.808 2 28.22 0.811 4 28.40 0.816 1 28.21 0.813 8 28.36 0.818 1 28.81 0.824 9 Parking Lot 24.45 0.836 8 24.66 0.840 4 25.25 0.854 3 25.40 0.854 8 25.57 0.859 3 25.53 0.860 5 River 29.25 0.793 1 29.22 0.793 5 29.40 0.796 2 29.34 0.796 6 29.29 0.796 0 29.39 0.796 8 Runway 31.25 0.830 4 31.15 0.831 0 31.56 0.835 2 31.85 0.839 2 31.26 0.832 4 31.76 0.840 8 S-Residential 31.57 0.865 8 31.64 0.868 7 31.87 0.868 9 31.74 0.868 3 31.85 0.868 2 31.91 0.870 5 Storage Tanks 32.71 0.896 5 32.95 0.897 4 33.16 0.899 9 33.03 0.898 8 33.14 0.901 0 33.36 0.902 2 Tennis Court 32.51 0.894 4 32.71 0.896 0 33.12 0.902 3 32.92 0.899 4 33.07 0.899 8 33.24 0.905 3 Average 29.92 0.840 8 30.00 0.842 0 30.34 0.847 6 30.25 0.845 8 30.32 0.847 2 30.45 0.850 9
Table 5. PSNR/SSIM of different methods on 30 types of RS scenes on AID test dataset for ×4 SR
View tableView in ArticleTable 5. PSNR/SSIM of different methods on 30 types of RS scenes on AID test dataset for ×4 SR
Class 遥感图像超分方法 TransENet HSENet HAUNet TTST TSFNet 本文 PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM Airport 29.23 0.818 2 29.22 0.814 7 29.32 0.816 3 29.29 0.815 4 29.38 0.819 5 29.47 0.821 1 Bare Land 36.20 0.862 6 36.21 0.862 3 36.40 0.865 7 36.20 0.864 4 36.18 0.864 2 36.22 0.868 3 Baseball Field 31.59 0.853 8 31.61 0.851 3 31.67 0.855 4 31.70 0.856 4 31.76 0.856 9 31.80 0.857 2 Beach 32.55 0.827 0 32.55 0.825 1 32.69 0.827 3 32.54 0.827 0 32.58 0.829 3 32.62 0.830 1 Bridge 31.63 0.854 4 31.61 0.853 2 31.77 0.855 6 31.68 0.851 6 31.79 0.854 4 31.82 0.856 7 Center 28.03 0.784 8 27.68 0.779 6 28.14 0.786 5 27.87 0.785 9 27.96 0.784 1 28.19 0.788 5 Church 24.51 0.703 8 24.38 0.698 9 24.58 0.705 4 24.47 0.700 5 24.59 0.707 8 24.60 0.708 7 Commercial 27.97 0.794 2 27.96 0.789 4 28.05 0.796 6 27.99 0.788 9 28.09 0.797 8 28.13 0.798 4 D-Residential 25.13 0.729 6 25.31 0.725 7 25.24 0.735 2 25.39 0.720 7 25.48 0.737 6 25.56 0.738 8 Desert 39.31 0.918 2 39.25 0.917 6 39.58 0.918 8 39.38 0.917 0 39.47 0.918 5 39.50 0.921 9 Farmland 34.58 0.864 5 34.46 0.862 2 34.72 0.866 5 34.45 0.859 9 34.63 0.864 0 34.68 0.869 6 Forest 28.56 0.705 9 28.68 0.703 0 28.61 0.711 9 28.67 0.702 9 28.70 0.704 8 28.72 0.708 1 Industrial 27.21 0.779 3 27.22 0.775 6 27.30 0.782 6 27.23 0.769 9 27.41 0.784 5 27.48 0.786 8 Meadow 32.66 0.748 4 32.67 0.745 9 33.02 0.747 4 32.79 0.746 2 32.86 0.748 6 32.92 0.749 6 M-Residential 28.45 0.754 4 28.56 0.750 0 28.55 0.756 4 28.60 0.754 7 28.65 0.756 9 28.75 0.758 2 Mountain 29.28 0.758 2 29.09 0.755 8 29.32 0.759 5 29.11 0.758 0 29.25 0.758 7 29.28 0.762 3 Park 28.01 0.770 2 28.06 0.768 4 28.08 0.771 5 28.06 0.769 6 28.17 0.772 8 28.23 0.773 5 Parking 26.40 0.835 5 26.30 0.823 4 26.63 0.836 8 26.45 0.829 6 26.84 0.837 2 26.92 0.838 8 Playground 30.30 0.832 0 30.12 0.826 2 30.16 0.834 1 30.15 0.831 1 30.40 0.834 4 30.48 0.835 9 Pond 30.53 0.823 1 30.43 0.821 3 30.60 0.824 3 30.48 0.821 8 30.54 0.824 0 30.57 0.824 9 Port 26.91 0.825 6 26.88 0.821 9 27.01 0.827 7 26.99 0.825 9 27.08 0.828 0 27.18 0.830 0 Railway Station 28.61 0.779 1 28.71 0.772 4 28.70 0.782 8 28.78 0.783 1 28.85 0.783 9 28.92 0.785 3 Resort 28.08 0.783 5 27.92 0.779 6 28.18 0.785 1 27.94 0.783 6 28.12 0.785 0 28.16 0.787 6 River 31.00 0.783 0 30.89 0.781 1 31.06 0.783 6 30.90 0.781 3 30.07 0.783 8 31.13 0.784 3 School 27.22 0.778 3 27.17 0.771 2 27.31 0.779 6 27.28 0.778 4 27.34 0.779 8 27.42 0.780 8 S-Residential 26.63 0.645 2 26.74 0.641 3 26.66 0.646 9 26.74 0.645 9 26.80 0.645 6 26.83 0.648 0 Square 29.39 0.804 5 29.16 0.800 7 29.52 0.806 6 29.18 0.803 5 29.45 0.806 0 29.58 0.808 8 Stadium 27.41 0.796 5 27.30 0.789 6 27.53 0.798 9 27.33 0.796 8 27.50 0.799 8 27.62 0.801 0 Storage Tanks 26.20 0.739 4 26.28 0.736 0 26.27 0.741 2 26.22 0.739 8 26.41 0.742 6 26.46 0.744 3 Viaduct 28.21 0.762 9 28.05 0.757 8 28.29 0.764 4 28.07 0.761 2 28.25 0.765 7 28.32 0.768 5 Average 29.38 0.790 9 29.32 0.786 7 29.49 0.792 4 29.40 0.789 0 29.50 0.792 6 29.59 0.794 5
Tools
Get Citation
Copy Citation Text
Yichuan YANG, Zhongqi MA, Xinyao ZHOU, Fujian ZHENG, Hong HUANG. Spatial adaptation and frequency fusion network for single remote sensing image super-resolution[J]. Optics and Precision Engineering, 2025, 33(8): 1238
Paper Information
Category:
Received: Dec. 16, 2024
Accepted: --
Published Online: Jul. 1, 2025
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20