Scene classification using high spatial resolution multispectral data

Abstract

Spectral imagery has traditionally been an important tool for terrain categorization (TERCAT). High-spatial resolution (8-meter), 4-color MSI data from IKONOS provide a new tool for scene classification. The utility of these data are studied for the purpose of classifying the Elkhorn Slough and surrounding wetlands in central California. The specific goal was to determine to what degree an existing classification map could be replicated using the 4-color imagery. The existing map was used as an input to a supervised classification process. Errors in that map required development of revised exemplar spectra sets, eliminating mixed classes. Classification was done using a spectral angle mapper and maximum likelihood classifier. Results were compared to the original classification map. Confusion matrix calculations showed agreement at the 10-20% level. This lack of agreement is attributed to errors in the original map at the relatively high resolution of IKONOS.