In hyperspectral images, some spectral bands suffer from low signal-to-noiseratio due to noisy acquisition and atmospheric effects, thus requiring robusttechniques for the unmixing problem. This paper presents a robust supervisedspectral unmixing approach for hyperspectral images. The robustness is achievedby writing the unmixing problem as the maximization of the correntropycriterion subject to the most commonly used constraints. Two unmixing problemsare derived: the first problem considers the fully-constrained unmixing, withboth the non-negativity and sum-to-one constraints, while the second one dealswith the non-negativity and the sparsity-promoting of the abundances. Thecorresponding optimization problems are solved efficiently using an alternatingdirection method of multipliers (ADMM) approach. Experiments on synthetic andreal hyperspectral images validate the performance of the proposed algorithmsfor different scenarios, demonstrating that the correntropy-based unmixing isrobust to outlier bands.
展开▼