Convex cone-based endmember extraction for hyperspectral imagery

Abstract

N-finder algorithm (N-FINDR) is a simplex-based fully abundance constrained technique which is operated on the original data space. This paper presents an approach, convex-cone N-FINDR (CC N-FINDR) which combines N-FINDR with convex cone data obtained from the original data so as to improve the N-FINDR in computational complexity and performance. The same convex cone approach can be also applied to simplex growing algorithm (SGA) to derive a new convex cone-based growing algorithm (CCGA) which also improves the SGA in the same manner as it does for NFINDR. With success in CC N-FINDR and CCGA a similar treatment of using convex cone can be further used to improve any endmember extraction algorithm (EEA). Experimental results are included to demonstrate advantages of the convex cone-based EEAs over EEAs without using convex cone.