IV
Unsupervised Hyperspectral Image Analysis
Unsupervised target analysis is one of principal strengths that hyperspectral imaging has edge over multispectral image processing. This is due to its high spectral resolution that allows users to uncover and reveal many unknown signal sources such as subtle material substances, subpixel targets, mixed constituent compositions, anomalies, etc. However, this advantage also comes at a price that target analysis must be performed by unsupervised means because such targets of interest generally cannot be identified by visual inspection or prior knowledge. Three specific applications are of particular interest in hyperspectral data exploitation: target discrimination, unsupervised target detection, and unsupervised target classification. In unsupervised target detection target, knowledge is not provided a priori in which case potential targets must be obtained directly from the data to be processed without prior knowledge. In this case, an issue is how to discriminate one detected target from another. As for unsupervised target classification it is more challenging because two main key issues, which are not encountered in unsupervised target detection, must be addressed. One is how to determine the number of targets of interest assumed to be in the data for classification. The other is how to find unknown target training samples for classification without prior knowledge. Part IV is included to address these issues. Three chapters are included ...
