Images of ground scenes have a tradeoff between spatial and spectral resolution. Sensors with fine spectral resolution sacrifice high spatial resolution. A hyperspectral image has tens to hundreds of bands of spectral information, which can potentially be used to increase understanding of the spatial content of a scene despite resolution limitations. While some pixels in an image could be identified as "trees" or "roads," for example, a number of pixels will be mixed, containing two or more material classes. The process called unmixing calculates the fractional presence of those predetermined materials. A new statistics-based stepwise unmixing routine has recently been developed that is an improvement over the traditional linear unmixing approach, as indicated by tests on synthetic data. This research has used quantitative tests on actual hyperspectral imagery to verify this hypothesis. In addition, t he stepwise procedure has been shown to out-perform a hierarchical linear unmixing routine, which is a more robust form of traditional unmixing.
Newland, Daniel, "Evaluation of stepwise spectral unmixing with HYDICE data" (1999). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus