Hyperspectral Unmixing With Spectral Variability Using a Perturbed Linear Mixing Model

摘要: Given a mixed hyperspectral data set, linear unmixing aims at estimating the reference spectral signatures composing data—referred to as endmembers—their abundance fractions and their number. In practice, identified endmembers can vary spectrally within given image thus be construed variable instances of endmembers. Ignoring this variability induces estimation errors that are propagated into procedure. To address issue, endmember consists from which estimated have been derived well with respect these references. This paper introduces new mixing model explicitly accounts for spatial variabilities. The parameters using an optimization algorithm based on alternating direction method multipliers. performance proposed is evaluated synthetic real data. A comparison state-of-the-art algorithms designed estimate allows interest solution appreciated.