EFFICIENT MAXIMUM-LIKELIHOOD CLASSIFICATION FOR IMAGING SPECTROMETER DATA SETS

A simplified maximum likelihood classification technique for handling remotely sensed image data is proposed which reduces, significantly, t he processing time associated with traditional maximum likelihood clas sification when applied to imaging spectrometer data, and copes with t he training of geographically small classes. Several wavelength subgro ups are formed from the complete set of spectral bands in the data, ba sed on properties of the global correlation among the bands. Discrimin ant values are computed for each subgroup separately and the sum of di scriminants is used for pixel labeling. Several subgrouping methods ar e investigated and the results show that a compromise among classifica tion accuracy, processing time, and available training pixels can be a chieved by using appropriate subgroup sizes.