“…The compared methods have the same compressive sampling scheme, which is based on the coded aperture snapshot spectral imaging system [10] and utilises the Bernoulli random matrix as the sensing matrix. The main difference among the compared methods lies in the sparse recovery, where the wavelet [10], the feature representation base [13], the cluster sparsity field [28], the inter‐spectral multi‐layered conditional random field [29], and multi‐spectral low‐rank structured dictionary learning [30] are utilised as the representation bases, respectively, in the framework of compressive sensing. In the following, the compared methods are named ‘Wavelet’, ‘FR’, ‘CSF’, ‘IS‐MCF’, and ‘MLSDL’, respectively, for short.…”