A Graph-Regularized Double-Path Interactive Spectral Super-Resolution Network for Hyperspectral Image Reconstruction

Deep learning has demonstrated outstanding potential for the spectral super-resolution (S2R) reconstruction of multispectral images (MSIs). However, it is still a challenge to alleviate spectral distortion during S2R reconstruction. Given the superiority of a graph, a graph-regularized double-path interactive S2R network (GDIS2Net) consisting of two parallel branches is proposed to reconstruct hyperspectral images (HSIs) from MSIs. An interactive residual module is carefully schemed as the backbone of the S2R network to facilitate the feature interaction between the two branches, while an enhanced residual module is constructed for further feature fusion. In addition, a new loss function considering the spectral continuity is proposed to optimize the proposed GDIS2Net. Experimental analyses show that the proposed GDIS2Net outperforms state-of-the-art methods on both simulated and real datasets.