Improved Spectral Unmixing of Hyperspectral Images Using Spatially Homogeneous Endmembers

Hyperspectral imaging is a new technique in remote sensing which provides image data at hundreds of spectral wave-lengths, thus allowing a very detailed characterization of the surface of the Earth (from an airborne or satellite platform). One of the most important challenges in hyperspectral imaging is to find an adequate pool of pure signature spectra of the materials present in the scene. These pure signatures are then used to decompose the scene into a set of so-called abundance fractions by means of a spectral unmixing algorithm, thus allowing a detailed analysis of the scene with sub-pixel precision. Most techniques available in endmember extraction literature rely on exploiting the spectral properties of the data alone. As a result, the search for endmembers in a scene is often conducted by treating the data as a collection of spectral measurements with no spatial arrangement. In this paper, we propose a novel strategy to incorporate spatial information into the traditional spectral-based endmember search process. Specifically, we propose to estimate, for each pixel vector in the scene, a scalar value which is used to weight the importance of the spectral information associated to each pixel in terms of its spatial context. The proposed methodology, which favours the selection of highly representative endmembers located in spatially homogeneous areas, is shown in this work to significantly improve several spectral-based endmember extraction algorithms available in the literature.