Infrared and Laser Engineering, Volume. 54, Issue 5, 20240592(2025)

Degradation-aware transformer for blind hyperspectral and multispectral image fusion(back cover paper·invited)

Xuheng CAO1,2,3, Xiaopeng HAO1,3, Yusheng LIAN4, Xuquan WANG2, and Xinbin CHENG2
Author Affiliations
  • 1Remote Sensing Calibration Laboratory, National Institute of Metrology, Beijing 100029, China
  • 2Institute of Precision Optical Engineering, School of Physics Science and Engineering, Tongji University, Shanghai 200092, China
  • 3Technology Innovation Center of Infrared Remote Sensing Metrology Technology, State Administration for Market Regulation, Beijing 100029, China
  • 4School of Printing and Packaging Engineering, Beijing Institute of Graphic Communication, Beijing 102600, China
  • show less
    Figures & Tables(8)
    Architecture of proposed (a) SpaDNet and (b) SpeDNet
    The proposed feature fusion network design diagram
    Architecture diagram of the proposed feature cross-fusion unit
    SAM error map of reconstructed images on (a), (b) CAVE and (c), (d) Harvard datasets
    The spectral band image and its zoom images reconstructed by ours and competing methods
    • Table 1. Comparison results of blind fusion performance of ten testing methods on CAVE and Harvard datasets

      View table
      View in Article

      Table 1. Comparison results of blind fusion performance of ten testing methods on CAVE and Harvard datasets

      CAVE datasetHarvard dataset
      MethodsPSNRSAMERGASSSIMPSNRSAMERGASSSIM
      HySure39.0122.330.660.98441.616.220.560.983
      LTMR31.4113.181.470.93040.225.580.890.970
      CNN-Fus34.8517.211.020.96036.6813.330.720.883
      HyMS42.434.850.580.98445.623.680.670.986
      FusFormer43.164.780.370.99748.012.990.300.994
      MoG-DCN43.174.780.340.99547.952.960.350.995
      DBSR39.405.250.510.99346.622.940.320.994
      SVPL40.605.980.590.99047.292.990.330.995
      UDALN38.696.290.580.99044.914.160.620.991
      Ours43.983.990.320.99748.712.840.300.997
    • Table 2. Ablation study of subspace loss

      View table
      View in Article

      Table 2. Ablation study of subspace loss

      Subspace lossPSNRSAMSSIM
      ×43.014.330.996
      43.983.990.997
    • Table 3. Computational analysis of deep-learning based methods on the Harvard dataset

      View table
      View in Article

      Table 3. Computational analysis of deep-learning based methods on the Harvard dataset

      FusFormerMoG-DCNDBSRSVPLUDALNOurs
      Flops0.16 T3.04 T3.60 T5.86 T78.38 G3.51 T
      Trainable parameter0.31 M9.98 M9.39 M8.0 M11953.01 M
      Time (Train/Infer)169.4 h/3.8 min46.7 h/1.7 s0/49.7 min4.5 h/1.8 s0/42.3 min0/41.8 min
    Tools

    Get Citation

    Copy Citation Text

    Xuheng CAO, Xiaopeng HAO, Yusheng LIAN, Xuquan WANG, Xinbin CHENG. Degradation-aware transformer for blind hyperspectral and multispectral image fusion(back cover paper·invited)[J]. Infrared and Laser Engineering, 2025, 54(5): 20240592

    Download Citation

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

    Category: Special issue—Hyperspectral technology and applications

    Received: Dec. 19, 2024

    Accepted: --

    Published Online: May. 26, 2025

    The Author Email:

    DOI:10.3788/IRLA20240592

    Topics