Spatial constraints on endmember extraction and optimization of per-pixel endmember sets for spectral unmixing

摘要

Fractional abundances predicted for a given pixel using spectral mixture analysis (SMA) are most accurate when only the spectral endmembers that comprise it are used, with larger errors occurring if inappropriate endmembers are included in the mixing process. Thus, in order to produce accurate results from spectral mixture analysis it is necessary to acquire representative endmember spectra of all image components and unmix each pixel using the appropriate endmember set for each pixel. In this paper we present an image endmember extraction algorithm which integrates spatial constraints in the search process and a spectral mixture algorithm designed to optimize the endmember set on a per-pixel basis. Implications on fractional abundances resulting from spectral unmixing analysis are then discussed.