A hybrid algorithm for subpixel detection in hyperspectral imagery

TitleA hybrid algorithm for subpixel detection in hyperspectral imagery
Publication TypeConference Papers
Year of Publication2004
AuthorsBroadwater J, Meth R, Chellappa R
Conference NameGeoscience and Remote Sensing Symposium, 2004. IGARSS '04. Proceedings. 2004 IEEE International
Date Published2004/09//
Keywordsalarm, algorithm;abundance, algorithm;Fully, algorithm;hybrid, algorithm;statistical, AMSD;Adaptive, analysis;structured, approximations;maximum, backgrounds;subpixel, constrained, detection;alarm, detection;target, Detector;FCLS, detector;hyperspectral, estimation;, estimation;emittance, identification;adaptive, imagery;reflectance, least, likelihood, Matched, matching;least, processing;geophysical, rate;generalized, ratio, signal, signatures;spectral, spectra;false, spectra;spectral, squares, subspace, systems;geophysical, techniques;image, tests;hybrid, unmixing

Numerous subpixel detection algorithms utilizing structured backgrounds have been developed over the past few years. These range from detection schemes based on spectral unmixing to generalized likelihood ratio tests. Spectral unmixing algorithms such as the Fully Constrained Least Squares (FCLS) algorithm have the advantage of physically modeling the interactions of spectral signatures based on reflectance/emittance spectroscopy. Generalized likelihood ratio tests like the Adaptive Matched Subspace Detector (AMSD) have the advantage of identifying targets that are statistically different from the background. Therefore, a hybrid detector based on both AMSD and FCLS was developed to take advantage of each detector's strengths. Results demonstrate that the hybrid detector achieved the lowest false alarm rates while also producing meaningful abundance estimates