A spatial compositional model (SCM) for linear unmixing and endmember uncertainty estimation

摘要

The normal compositional model (NCM) has been extensively used inhyperspectral unmixing. However, most of the previous research has focused onestimation of endmembers and/or their variability. Also, little work hasemployed spatial information in NCM. In this paper, we show that NCM can beused for calculating the uncertainty of the estimated endmembers with spatialpriors incorporated for better unmixing. This results in a spatialcompositional model (SCM) which features (i) spatial priors that forceneighboring abundances to be similar based on their pixel similarity and (ii) aposterior that is obtained from a likelihood model which does not assume pixelindependence. The resulting algorithm turns out to be easy to implement andefficient to run. We compared SCM with current state-of-the-art algorithms onsynthetic and real images. The results show that SCM can in the main providemore accurate endmembers and abundances. Moreover, the estimated uncertaintycan serve as a prediction of endmember error under certain conditions.