Recent Advances and Challenges in Automatic Hyperspectral Endmember Extraction

The advancements in hyperspectral remote sensing are increasing continuously and recording a wealth of spatial as well as spectral information about an object, but resulting high volume of data. Analysis and classification of this high volume hyperspectral data needs a ground truth data or spectral library or image based endmembers which assist to unmix the mixed pixels and map their spatial distribution. Till date, though several hyperspectral endmember extraction algorithms have been proposed, every algorithm has its own limitations. The perfect endmember extraction algorithm would find unique spectra with no prior knowledge. This paper discusses the recent improvements and challenges in hyperspectral endmember extraction. The algorithms evaluated includes PPI, NFINDR, FIPPI and ATGP. The experiments are performed on the subset of Hyperion and AVIRIS_NG datasets.