Unmixing analysis based on multiscale segmentation

Unmixing plays an important role in hyperspectral image processing and the wide range of application of hyperspectral data. Fully-automated techniques that take into account the spatial and spectral information captured by the hyperspectral sensor are required. This paper describes a new approach for unmixing analysis of hyperspectral imagery using a novel multiscale segmentation technique for the estimation of both the number of endmembers and their spectral signatures. The proposed approach uses the spatial and spectral information of hyperspectral imagery. In addition, the described method seeks to model the natural variability of the materials using multiple spectral to build endmember classes. Experimental results using an image from SOC 700 hyperspectral imagery are presented.