Endmember extraction and classification of tropical trees (India) using SFF & SAM algorithm

Abstract

Present study investigates the endmember extraction and classification of tropical trees (India) using Spectral Feature Fitting (SFF) & Spectral Angle Mapper (SAM) algorithms. Space-borne Hyperion data was acquired for two different dates (October, 2006 & January, 2011). Endmembers were picked up from highest percentage occupancy of a class. Classification was performed on a combination of bands coming from specific absorptive features of endmember spectra (VIS (458nm-702nm) and SWIR-I (1548nm-1780nm)). SFF & SAM were applied on two identified band combinations a) 25bands (VIS) & b) 49 bands (VIS+SWIR-I). SFF was performed on Continuum Removed (CR) spectra and SAM on Continuum Intact (CI) spectra. Overall accuracy (OAA) levels were different for the two band combinations, two algorithms and also for the two images. For October data, SFF gave 61% and 49% OAA using VIS+SWIR-1 bands and VIS bands respectively. The SAM classifier performed better for both the selected spectral regions (70% and 67% respectively). SAM showed better performance with 68% using VIS bands and SFF gave good OAA (56%) with VIS+SWIR-1 band combination for January data. Results indicate that OAA defers with changes in the phenology of vegetation cover. Endmember spectra worked well in classifying images coming from two different acquisition dates.