JR2net: A Joint Non-Linear Representation and Recovery Network for Compressive Spectral Imaging

Abstract

Compressive spectral imaging (CSI) senses 3D spatial-spectral information through 2D coded projections. Reconstruction methods conventionally rely on handcrafted priors or end-to-end deep learning. JR2net proposes a joint non-linear representation and recovery network that integrates learned low-dimensional representations directly into the recovery optimization process, substantially boosting reconstruction fidelity and speed.