Unsupervised unmixing of hyperspectral imagery using a multiscale representation.
Abstract
This paper presents an unsupervised unmixing approach that takes advantage of multiscale representation based on nonlinear diffusion to extract the spectral endmembers from an image. The main features of the proposed approach are the use of spatial information and the avoidance of matrix rank estimation to determine the number of endmembers in the image. Multiscale representation builds a family of smoothed image where locally spectrally uniform regions can be identified. The multi-scale representation is extracted using a nonlinear diffusion PDE. Locally homogeneous regions are identified by taking advantage of an algebraic multigrid method used to solve the PDE. Representative spectral signatures for each region are extracted and then clustered to build spectral endmember classes. these classes represent the different spectral components of the image as well as their spectral variability. The number of spectral endmember classes is estimated using the Davies and Bouldin validity index. Experimental results using the AVIRIS image of Fort A.P. Hill are presented which show the potential of the proposed method.