Maximum likelihood classification combined with spectral angle mapper algorithm for high resolution satellite imagery

Abstract

The generation of precise land cover classification maps is an important application of high resolution satellite multispectral imagery. In this study, Spectral Angle Mapper algorithm (SAM) was used to extract the spectral characteristics from multispectral imagery. The spectral angle between neighbouring pixels was calculated. The distribution of spectral characteristics was derived from the average and variance of the calculated spectral angle in a 3×3 window of the image. The extracted spectral characteristics were combined with original multispectral imagery, and the data were classified by the maximum likelihood method. This approach was applied to Quickbird multispectral imagery. The extracted spectral characteristics highlighted boundaries between different types of land cover. The method proposed in this study exhibits an increase in overall classification accuracy relative to the original maximum likelihood method.