ABSTRACT
This chapter deals with a discussion of the problems posed by the classification of data with high spectral resolution, that is, hyperspectral imagery, including an introduction to linear spectral un-mixing and the derivation of algorithms for linear and kernel anomaly detection. It discusses postclassification processing methods to improve classification results on the basis of contextual information. The chapter examines the adaptive boosting technique, applying it in particular to improve the generalization accuracy of neural network classifiers. The supervised classification case is somewhat different, since reference is always being made to a—generally quite small—subset of labeled training data. Assuming that sufficient labeled data are available for some to be set aside for test purposes, test data can be used to make an unbiased estimate of the misclassification rate of a trained classifier, that is, the fraction of new data that will be incorrectly classified.
