Hyperspectral Data Processing: Algorithm Design and Analysis

18

Pixel Extraction and Information

Because of very high spectral resolution provided by a hyperspectral imager, a single hyperspectral image pixel vector can now unveil subtle and crucial information for data analysis that a traditional image pure-based pixel cannot. Unfortunately, much of such rich information is presumably unknown and cannot be identified a priori. This chapter investigates the issue of how to extract pixel information from an exploitation viewpoint. In order to facilitate pixel information analysis, four types of pixels of interest, pure pixel, mixed pixel, anomalous pixel, and homogeneous pixel, are defined. A pure pixel is a pixel whose spectral signature is completely represented by a single-material substance as opposed to a mixed pixel whose spectral signature is composed of more than one material substance. A homogeneous pixel can be defined as a pixel whose spectral signature remains nearly constant subject to small variations within its surroundings in contrast to an anomalous pixel whose signature is spectrally distinct from the signatures of its neighboring pixels. Therefore, a homogeneous pixel can be considered opposite to an anomalous pixel, while a pure pixel is an opposite of a mixed pixel. On one end, pure and mixed pixels can be dealt with from a single-pixel point of view. On the other end, analysis of homogeneous and anomalous pixels must take into account the surrounding pixels within their neighborhoods, that is, their neighboring pixels. ...