7.2 Convex Geometry-Based Endmember Extraction

Due to the nature of a mixed sample vector in a linear mixing model, convex geometry has been used as a major criterion behind EEA design. As a simple illustrative example, assume that e1 and e2 are two endmembers. Any data sample vector x lying in a line segment connecting these two endmembers as end points can be expressed as a point linearly mixed by e1 and e2 and described in the form of img, with img based on convexity. For three endmembers e1, e2, and e3, a data sample vector that is linearly mixed by these three endmembers must lie in a triangle with e1, e2, and e3 as its three vertices. Similarly, a data sample vector within a triangular pyramid/tetrahedron, square-pyramid, and p-vertex simplex is also considered to be linearly mixed by its four vertices, five vertices, and p vertices, respectively, all of which are considered as endmembers. The OP and simplex volume are two principal criteria to materialize the concept of convex geometry.

7.2.1 Convex Geometry-Based Criterion: Orthogonal Projection

PPI is a popular technique widely used in endmember extraction due to its availability in the ENVI software provided by the Research Systems. Also due to its propriety and limited publication, its detailed implementation has never been made ...

Get Hyperspectral Data Processing: Algorithm Design and Analysis now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.