Supervised versus unsupervised methods for classification of coasts and river corridors from airborne remote sensing

Abstract

Maximum likelihood classification (MLC) is compared with unsupervised classification (USC) for operational cover mapping from airborne imagery. Data from the Compact Airborne Spectrographic Imager (CASI) were clustered into a large number of unsupervised subclasses then amalgamated into target classes. For extensive operational classification of coastal imagery, USC proved an acceptable alternative to MLC due to the wide spectral variation between surface classes. However, for river corridors containing a variety of vegetated surfaces with similar reflectances, USC did not produce consistent results.