Hyperspectral Data Processing: Algorithm Design and Analysis

16.5 Conclusions

A simplest unsupervised hyperspectral target analysis is to use a hyperspectral measure to perform target signature discrimination. When there is prior knowledge available, a hyperspectral measure can take advantage of it to be further used to accomplish tasks other than discrimination. For example, if there is a training data set for class membership available, a hyperspectral measure can work as a classifier either as a hard decision-made classifier or a soft decision-made quantifier. Moreover, if there is a database or spectral library available, a hyperspectral measure can be performed to be used for signature verification or identification. This chapter derives two categories of hyperspectral measures, signature vector-based hyperspectral measures and correlation matrix-weighted hyperspectral measures. While the former is generally considered as spectral similarity measures commonly used in remote sensing community, the latter has been used as various forms as detectors, classifiers, or identifiers due to the fact that they can take into account the correlation among data sample vectors to be used for various tasks. Such sample correlation information can be characterized by two types of information, a priori information and a posteriori information which can be used to design correlation matrix-weighted hyperspectral measures. As examples, if a priori information is provided by a set of training data for class membership, correlation matrix-weighted hyperspectral ...