Linear-quadratic and polynomial Non-Negative Matrix Factorization; application to spectral unmixing

In this article, we present a source separation method for linear-quadratic models. This class of mixing models is encountered in various real applications, such as hyperspectral unmixing for urban environments. Linear-quadratic mixing models are less studied in the literature than linear ones but there exist some methods for handling them, essentially Bayesian or based on Independent Component Analysis.