Incorporation of spatial constraints into spectral mixture analysis of remotely sensed hyperspectral data

Spectral mixture analysis is an important technique to analyze remotely sensed hyperspectral data sets. This approach involves the separation of a mixed pixel into its pure components or endmember spectra, and the estimation of the abundance value for each endmember. Several techniques have been developed for extraction of spectral endmembers and estimation of fractional abundances. However, an important issue that has not been yet fully accomplished is the incorporation of spatial constraints into endmember extraction and, particularly, fractional abundance estimation. Another relevant topic is the use of nonlinear versus linear mixture models, which can be unconstrained or constrained in nature. Here, the constraints refer to non-negativity and sum to unity of estimated fractional abundances for each pixel vector. In this paper, we investigate the impact of including spatial and abundance-related constraints in spectral mixture analysis of remotely sensed hyperspectral data sets. For this purpose, we discuss the advantages that can be obtained after including spatial information in techniques for endmember extraction and fractional abundance estimation, using a database of synthetic hyperspectral scenes with artificial spatial patterns generated using fractals, and a real hyperspectral scene collected by NASA's airborne visible infra-red imaging spectrometer (AVIRIS).