Spectral Variability Augmented Sparse Unmixing of Hyperspectral Images

Abstract

Spectral unmixing (SU) expresses the mixed pixels existed in hyperspectral\nimages as the product of endmember and abundance, which has been widely used in\nhyperspectral imagery analysis. However, the influence of light, acquisition\nconditions and the inherent properties of materials, results in that the\nidentified endmembers can vary spectrally within a given image (construed as\nspectral variability). To address this issue, recent methods usually use a\npriori obtained spectral library to represent multiple characteristic spectra\nof the same object, but few of them extracted the spectral variability\nexplicitly. In this paper, a spectral variability augmented sparse unmixing\nmodel (SVASU) is proposed, in which the spectral variability is extracted for\nthe first time. The variable spectra are divided into two parts of intrinsic\nspectrum and spectral variability for spectral reconstruction, and modeled\nsynchronously in the SU model adding the regular terms restricting the sparsity\nof abundance and the generalization of the variability coefficient. It is noted\nthat the spectral variability library and the intrinsic spectral library are\nall constructed from the In-situ observed image. Experimental results over both\nsynthetic and real-world data sets demonstrate that the augmented decomposition\nby spectral variability significantly improves the unmixing performance than\nthe decomposition only by spectral library, as well as compared to\nstate-of-the-art algorithms.\n