• Acta Photonica Sinica
  • Vol. 51, Issue 11, 1110003 (2022)
Xueyuan GUAN, Wei HU*, and Heng FU
Author Affiliations
  • State Key Laboratory of Transient Physics,Nanjing University of Science and Technology,Nanjing 210094,China
  • show less
    DOI: 10.3788/gzxb20225111.1110003 Cite this Article
    Xueyuan GUAN, Wei HU, Heng FU. Remote Sensing Image Denoising Algorithm with Multi-receptive Field Feature Fusion and Enhancement[J]. Acta Photonica Sinica, 2022, 51(11): 1110003 Copy Citation Text show less
    MRFENet structure diagram
    Fig. 1. MRFENet structure diagram
    Structure diagram of global feature extraction module
    Fig. 2. Structure diagram of global feature extraction module
    Examples of NWPU-RESISC45 dataset
    Fig. 3. Examples of NWPU-RESISC45 dataset
    Example of different noise intensity datasets
    Fig. 4. Example of different noise intensity datasets
    Example of RSSCN7 datasets and real noise datasets
    Fig. 5. Example of RSSCN7 datasets and real noise datasets
    Performance curves under different parameters
    Fig. 6. Performance curves under different parameters
    Loss curve chart
    Fig. 7. Loss curve chart
    Example of denoising results of different algorithms(σ=15)
    Fig. 8. Example of denoising results of different algorithms( σ=15)
    Example of denoising results of different algorithms(σ=20)
    Fig. 9. Example of denoising results of different algorithms(σ=20)
    Example of denoising results of different algorithms(σ=35)
    Fig. 10. Example of denoising results of different algorithms(σ=35)
    Example of denoising results of different algorithms(σ=50)
    Fig. 11. Example of denoising results of different algorithms(σ=50)
    Denoising results of various methods under different noise intensity
    Fig. 12. Denoising results of various methods under different noise intensity
    Example of real noise denoising result
    Fig. 13. Example of real noise denoising result
    DatasetMethodsPSNR(dB)/SSIM σ2 = 15PSNR(dB)/SSIM σ2 = 20PSNR(dB)/SSIM σ2 = 35PSNR(dB)/SSIM σ2 = 50PSNR(dB)/SSIM Hybrid noise
    NWPU-RESISC45NLM29.315/0.863 725.469/0.811 722.187/0.723 120.142/0.625 724.012/0.752 3
    BM3D29.588/0.878 026.352/0.832 524.732/0.769 522.025/0.710 525.131/0.793 1
    DnCNN30.732/0.924 128.565/0.850 327.182/0.833 725.149/0.789 727.973/0.839 0
    RIDNet30.893/0.924 529.438/0.875 127.679/0.841 525.493/0.796 328.512/0.852 6
    REDJ31.096/0.925 630.635/0.891 728.031/0.849 925.592/0.805 129.467/0.885 2
    MRFENet(ours)31.478/0.944 730.833/0.894 528.622/0.854 726.036/0.812 729.499/0.890 3
    RSSCN7NLM28.078/0.879 825.377/0.789 524.165/0.721 520.575/0.601 924.537/0.761 7
    BM3D28.842/0.886 826.174/0.812 424.691/0.760 321.362/0.673 525.882/0.790 1
    DnCNN31.949/0.929 728.051/0.843 726.322/0.839 225.033/0.819 327.256/0.840 3
    RIDNet32.340/0.943 428.866/0.889 328.149/0.847 225.474/0.822 928.481/0.867 4
    REDJ32.290/0.950 829.472/0.891 028.228/0.858 326.163/0.830 529.022/0.877 1
    MRFENet(ours)32.434/0.958 729.636/0.892 628.386/0.862 726.635/0.849 829.255/0.832 5
    Table 1. Quantitative results of different noise intensities under each method
    DatasetEvaluationNoisyNLMBM3DDnCNNRIDNetREDJMRFENet(ours)

    Washington DC

    mall

    NIQE↓7.978 78.439 39.595 19.405 610.329 412.738 58.138 5
    BRISQUE↓43.804 850.605 843.456 343.457 843.455 543.458 442.458 2
    PIQE↓39.366 083.656 448.993 338.598 640.470 745.659 737.056 6
    Table 2. Quantitative results of real noise images under each method
    ModuleGFECAMStep feature fusionPSNR(dB)/SSIM
    Experiment××28.563/0.870 2
    ××28.427/0.863 7
    ××28.556/0.869 3
    ×29.355/0.881 4
    ×29.165/0.873 7
    ×29.545/0.882 1
    29.636/0.892 6
    Table 3. Quantitative results of ablation experiments
    Xueyuan GUAN, Wei HU, Heng FU. Remote Sensing Image Denoising Algorithm with Multi-receptive Field Feature Fusion and Enhancement[J]. Acta Photonica Sinica, 2022, 51(11): 1110003
    Download Citation