Spatial-spectral unmixing using fuzzy local information

Hyperspectral unmixing estimates the proportions of materials represented within a spectral signature. The over whelming majority of hyperspectral unmixing algorithms are based entirely on the spectral signatures of each individual pixel and do not incorporate the spatial information found in a hyperspectral data cube. In this work, a spectral unmixing algorithm, the Local Information Proportion estimation (LIP) algorithm, is presented. The proposed LIP algorithm incorporates spatial information while determining the proportions of materials found within a spectral signature. Spatial information is incorporated through the addition of a spatial term that regularizes proportion value estimates based on the weighted proportion values of neighboring pixels. Results are shown in the AVIRIS Indian Pines hyperspectral data set.