16
Hyperspectral Measures
One simplest and easiest means to conduct unsupervised target analysis is to use hyperspectral measures for target discrimination detection, classification, recognition, and identification. Many hyperspectral measures have been studied in the literature, particularly in Chang (2003a, Chapter 2). They are primarily designed to measure spectral similarity among signatures for the purpose of detection, discrimination, classification, and identification. This chapter revisits several commonly used signature vector-based spectral similarity measures and further generalizes signature vector-based hyperspectral measures to sample correlation-weighted hyperspectral measures by including a weighting correlation matrix into a signature vector-based spectral measure so as to improve its performance. The idea of such generalization is similar to that using a weighting matrix to extend linear spectral mixture analysis (LSMA) to weighted abundance-constrained LSMA (WAC-LSMA) developed in Chapter 14.
