MFFNet: Multidomain Feature Fusion Network for Hyperspectral Pansharpening

Hyperspectral (HS) pansharpening is designed to fuse high spatial resolution panchromatic (PAN) images with low spatial resolution hyperspectral (LRHS) images to generate high spatial resolution hyperspectral (HRHS) images. Because of insufficient consideration of the interference caused by modal differences in spectral and spatial features during feature fusion, most deep learning-based methods suffer from spectral and spatial distortions in the fusion results. To address these issues, a multidomain feature fusion network (MFFNet) is proposed to perform fine-grained extraction and fusion of multiscale spatial-spectral features from PAN and HS images. A dual U-Net model is constructed to extract and fuse multiscale features from PAN and HS images, with a spatial-curvature feature enhancement module (SpaFEM) and a spectral-curvature feature enhancement module (SpeFEM) designed to improve curvature features. Moreover, a low-rank feature cross-attention module (LFCM) based on matrix factorization is introduced to enhance and complement the PAN and HS image features through cross-modal fusion in a low-dimensional space. Extensive experiments across four datasets are used to demonstrate that the proposed MFFNet outperforms several state-of-the-art methods. The code is available at https://doi.org/10.57760/sciencedb.010ns.