Hyperspectral Image Reconstruction Using a Deep Spatial-Spectral Prior

Regularization is a fundamental technique to solve an illposed optimizationproblem robustly and is essential to reconstruct compressive hyperspectralimages. Various handcrafted priors have been employed as a regularizer but are often insufficient to handle the wide variety of spectra of natural hyperspectral images, resulting in poor reconstruction quality. Moreover the prior-regularizedoptimization requires manual tweaking of its weight parameters to achieve a balance between the spatialand spectral fidelity of result images. In thispaper we presenta novel hyperspectralimage reconstructionalgorithm that substitutes the traditionalhand-craftedpriorwith a data-drivenprior, based on an optimization-inspirednetwork. Our method consists of two main parts: First, we learn a novel datadriven priorthat regularizes the optimizationproblem with a goal to boost the spatial-spectralfidelity.Our data-driven prior learns both local coherence and dynamic characteristics of natural hyperspectral images. Second, we combine our regularizerwith an optimization-inspirednetwork to overcome the heavy computation problem in the traditional iterative optimization methods. We learn the complete parameters in the network through end-to-end training, enabling robust performance with high accuracy. Extensive simulation and hardware experiments validate the superiorperformance of our method over the state-of-theartmethods.