LAUSR

This paper proposes a noise-resistant coded aperture snapshot spectral imaging (CASSI) reconstruction algorithm based on a spectral awareness network (SANet). The method maps the snapshot compressed measurements to a panchromatic (PAN) image as an auxiliary learning task for reconstructing hyperspectral images. The PAN images reconstructed by the network are extracted to provide spatial detail information, which is then fed into the CASSI reconstruction network to enhance its regularization capability. Experiments show that when extra Gaussian or Poisson noise is added, HSIs reconstructed by the SANet outperform other state-of-the-art (SOTA) methods in terms of the structural similarity index (SSIM) and spectral angle mapper (SAM), proving its high robustness in reconstruction capabilities.