Advances in Hyperspectral Image Processing Techniques

8Endmember Finding in Compressively Sensed Band Domain

Chein‐I Chang^1,2 and Adam Bekit¹

¹Remote Sensing Signal and Image Processing Laboratory, Department of Computer Science and Electrical Engineering, University of Maryland Baltimore County, Baltimore, MD, USA

²Center for Hyperspectral Imaging in Remote Sensing (CHIRS), Information and Technology College, Dalian Maritime University, Dalian, China

8.1 Introduction

Endmembers are pure signatures generally used to identify spectral classes in hyperspectral imagery [1, 2]. As a result, it is a fundamental task prior to spectral unmixing. Unfortunately, finding endmembers is a very challenging since it must be carried out in an unsupervised manner due to the fact that endmembers are generally unknown and relatively small targets which cannot be identified by visual inspection or prior knowledge. Despite that many approaches have been developed and reported in the literature, there are several issues. One is how to determine the number of endmembers to be found, p, which is generally much smaller than the total number of spectral bands, L. Fortunately, a recently developed concept, virtual dimensionality (VD) [3], can be used for this purpose. Once the p is determined, a following issue is finding a desired set of p endmembers. One of most widely used approaches is simplex volume (SV)‐based methods, which considers p endmembers as vertices to form a p‐vertex simplex and makes use of its SV as a criterion for finding a desired set ...