Geometric mixture analysis of imaging spectrometry data

Linear spectral mixture analysis, or unmixing, of imaging spectrometry data is essentially a geometry problem. Finite spatial resolution and natural heterogeneity conspire to make spectral mixing inherent in all imaging spectrometry data. The concepts of affine, convex and projective geometries provide a natural framework for understanding spectral mixing and tools for unraveling it. It is possible to automatically derive the number of mixing endmembers, estimates of their pure spectra and maps of their apparent surface abundances using only the mixed, observed data. Pure pixels are not required for the process.<
>