Laser & Optoelectronics Progress, Volume. 62, Issue 6, 0628003(2025)
Fusion Multiscale Receptive Field and Multilevel Hybrid Transformer for Super-Resolution Reconstruction of Remote Sensing Images
Figures & Tables(10)
Fig. 5. Runtime of different algorithms on PatternNet, WHU-RS19, AID, and NWPU-RESISC45 datasets
Table 1. Comparison of average PSNR of different models on different datasets
View tableTable 1. Comparison of average PSNR of different models on different datasets
Dataset Scale Bicubic SRGAN ESRGAN CAL_GAN SAFMN SRFormerV2 SRTrans GAN Proposed algorithm PatternNet ×2 29.01 36.91 37.29 37.10 36.58 37.50 37.76 37.99 ×3 26.03 32.55 33.45 33.15 31.32 33.65 33.76 34.11 ×4 24.34 30.16 30.87 30.55 29.74 31.10 31.23 31.67 AID ×2 29.43 38.37 38.75 38.45 37.25 38.89 39.44 39.55 ×3 26.82 35.02 35.62 35.38 33.73 35.80 36.33 36.52 ×4 23.34 33.18 33.68 33.43 31.96 33.83 34.35 34.73 WHU-RS19 ×2 28.59 37.15 37.66 37.35 36.57 37.89 38.75 39.06 ×3 24.55 35.84 36.46 36.20 33.81 36.60 37.80 38.08 ×4 22.96 33.94 34.46 34.15 31.63 34.65 35.71 36.08 NWPU-RESISC45 ×2 28.61 38.31 38.67 38.45 37.04 38.72 38.81 38.74 ×3 25.73 35.52 36.10 35.65 32.28 36.35 35.07 37.15 ×4 23.25 32.09 32.62 32.45 30.98 32.87 33.73 34.86 UCMERCED ×2 29.13 37.14 37.55 37.25 36.91 37.68 37.92 38.69 ×3 24.81 34.61 35.22 34.80 34.36 35.38 34.94 37.33 ×4 22.18 32.55 32.89 32.70 32.51 33.15 32.24 34.94
Table 2. Comparison of average SSIM of different models on different datasets
View tableTable 2. Comparison of average SSIM of different models on different datasets
Dataset Scale Bicubic SRGAN ESRGAN CAL_GAN SAFMN SRFormer V2 SRTrans
GAN
Proposed algorithm PatternNet ×2 0.856 0.960 0.965 0.962 0.937 0.968 0.970 0.972 ×3 0.794 0.906 0.922 0.918 0.883 0.928 0.932 0.935 ×4 0.737 0.848 0.885 0.881 0.814 0.892 0.893 0.897 AID ×2 0.812 0.902 0.993 0.991 0.892 0.995 0.996 0.998 ×3 0.762 0.957 0.960 0.954 0.917 0.964 0.964 0.967 ×4 0.713 0.908 0.928 0.924 0.874 0.932 0.935 0.936 WHU-RS19 ×2 0.801 0.944 0.942 0.934 0.920 0.948 0.951 0.954 ×3 0.742 0.897 0.902 0.896 0.861 0.911 0.917 0.922 ×4 0.689 0.926 0.968 0.964 0.898 0.975 0.979 0.988 NWPU-RESISC45 ×2 0.797 0.916 0.962 0.953 0.841 0.966 0.970 0.971 ×3 0.716 0.899 0.928 0.922 0.813 0.934 0.954 0.965 ×4 0.669 0.946 0.977 0.972 0.874 0.981 0.981 0.983 UCMERCED ×2 0.813 0.944 0.953 0.941 0.914 0.957 0.963 0.969 ×3 0.736 0.883 0.922 0.916 0.843 0.928 0.930 0.932 ×4 0.621 0.923 0.973 0.968 0.871 0.976 0.977 0.979
Table 3. Comparison of average FSIM of different models on different datasets
View tableTable 3. Comparison of average FSIM of different models on different datasets
Dataset Scale Bicubic SRGAN ESRGAN CAL_GAN SAFMN SRFormer V2 SRTrans GAN Proposed algorithm PatternNet ×2 0.861 0.994 0.996 0.995 0.988 0.996 0.997 0.997 ×3 0.852 0.99 0.994 0.992 0.989 0.995 0.994 0.996 ×4 0.834 0.981 0.988 0.985 0.976 0.991 0.992 0.993 AID ×2 0.824 0.904 0.910 0.908 0.989 0.912 0.942 0.986 ×3 0.814 0.901 0.904 0.902 0.896 0.906 0.902 0.903 ×4 0.798 0.892 0.897 0.895 0.887 0.899 0.896 0.899 WHU-RS19 ×2 0.832 0.912 0.917 0.914 0.907 0.919 0.913 0.915 ×3 0.822 0.906 0.910 0.908 0.902 0.912 0.912 0.913 ×4 0.806 0.898 0.905 0.901 0.894 0.909 0.907 0.909 NWPU-RESISC45 ×2 0.841 0.921 0.931 0.926 0.913 0.935 0.944 0.947 ×3 0.832 0.914 0.923 0.919 0.901 0.926 0.927 0.931 ×4 0.809 0.906 0.921 0.917 0.895 0.929 0.933 0.937 UCMERCED ×2 0.827 0.917 0.940 0.935 0.903 0.952 0.961 0.974 ×3 0.804 0.911 0.933 0.929 0.891 0.936 0.936 0.938 ×4 0.796 0.904 0.928 0.923 0.874 0.938 0.940 0.942
Table 4. Results of generator network ablation experiment
View tableTable 4. Results of generator network ablation experiment
Model setting PSNR /dB SSIM Generator network 37.99 0.972 Remove multi-scale feature extraction module 33.52 0.791 Remove dynamic feature fusion module 34.83 0.835 Remove MSHT module 35.15 0.897
Table 5. Results of generator network ablation experiment
View tableTable 5. Results of generator network ablation experiment
Model setting PSNR /dB SSIM Discriminator network 37.99 0.972 Remove multi-scale convolutional discriminator 34.60 0.811 Remove global transformer discriminator 35.49 0.867 Remove hierarchical feature discriminator 36.22 0.883
Tools
Get Citation
Copy Citation Text
Bo Li, Lingyun Kong, Mingwei Zhao, Xinyu Liu. Fusion Multiscale Receptive Field and Multilevel Hybrid Transformer for Super-Resolution Reconstruction of Remote Sensing Images[J]. Laser & Optoelectronics Progress, 2025, 62(6): 0628003
Paper Information
Category: Remote Sensing and Sensors
Received: Dec. 23, 2024
Accepted: Jan. 20, 2025
Published Online: Mar. 10, 2025
The Author Email: Lingyun Kong (konglingyun@xijing.edu.cn), Xinyu Liu (liuxinyu@huanghuai.edu.cn)
CSTR:32186.14.LOP242482
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20