ABSTRACT
Spectral unmixing (SU) of hyperspectral images (HSIs) has been in the spotlight of both research and applications during the recent years. Abundance estimation algorithms hinge on the assumption that the underlying mixing process is properly modeled. Along those lines, the linear mixing model (LMM) has been widely adopted by numerous unmixing algorithms. This chapter presents a new unmixing algorithm that imposes concurrently the two aforementioned constraints, that is, sparsity and low-ranknes. It formulates unmixing problem as a simultaneously sparse and low-rank matrix estimation problem with the corresponding abundance matrix constrained to be simultaneously sparse and low-rank. The chapter demonstrates the effectiveness of the proposed algorithm on both simulated and real data experiments. To better illustrate the advantages of the proposed algorithm in terms of estimation accuracy, the chapter compares it with three state-of-the-art unmixing algorithms, namely the constrained sparse unmixing by variable splitting and augmented Lagrangian algorithm, the nonnegatively constrained joint-sparse method, and the fast Bayesian inference iterative conditional expectations.
