Spatially-smooth piece-wise convex endmember detection

An endmember detection and spectral unmixing algorithm that uses both spatial and spectral information is presented. This method, Spatial Piece-wise Convex Multiple Model Endmember Detection (Spatial P-COMMEND), autonomously estimates multiple sets of endmembers and performs spectral unmixing for input hyperspectral data. Spatial P-COMMEND does not restrict the estimated endmembers to define a single convex region during spectral unmixing. Instead, a piece-wise convex representation is used that can effectively represent non-convex hyperspectral data. Spatial P-COMMEND drives neighboring pixels to be unmixed by the same set of endmembers encouraging spatially-smooth unmixing results.