Creating and sharing knowledge for telecommunications

Mixture of Dirichlet sources applied to hyperspectral unmixing

Nascimento, J. ; Bioucas-Dias, J.

Mixture of Dirichlet sources applied to hyperspectral unmixing, Proc EARSEL Symp., Oporto, Portugal, Vol. -, pp. 0 - 0, June, 2005.

Digital Object Identifier:

Abstract
Linear unmixing decomposes a hyperspectral image into a collection of reflectance spectra, called endmember
signatures, and a set of abundance fractions vectors (one vector per pixel). This paper introduces a new method to unmix hyperspectral data, where abundance fractions are modelled as Dirichlet sources.
This model forces abundance fractions to be non-negative and to have constant sum on each pixel. The
mixing matrix is inferred by an expectation-maximization (EM) type algorithm. The performance of the method is illustrated using simulated and real data.