Automatic Labelling of End Members Extracted from Hyperspectral Data

ISRO has launched Hyperspectral Satellite HYSIS on November 29 2018 carrying two sensors one covering VNIR and the other covering SWIR. VNIR sensor has spectral resolution of 9.42 nm in the wavelength range of 424 to 942 nm, providing 60 bands. SWIR sensor has spectral resolution of 6.25 nm in the spectral wavelength range of 881 to 2473 nm, providing 256 bands total of 316 bands. Large number of bands provides detailed spectral information about various terrestrial features, like, vegetation species, mineralogy, etc. An experienced analyst can identify all the classes with in the image manually and use them for classification. But for a user with limited understanding of hyperspectral remote sensing it is not easy. To overcome this difficulty, several approaches for automatic identification of endmembers or pure pixels have been developed over the years. However, labelling of the endmembers has to be carried out by the analyst manually. This paper describes methodology for automatic labelling of endmembers. This is achieved as a two-step procedure. Firstly, end members which are pure pixels/samples are automatically extracted from input Hyperspectral Image. The derived endmembers are matched with library spectra for assigning the labels. The labeled end members can be used as input for generating full or sub pixel classified outputs.