Improving the classification precision of spectral angle mapper algorithm

The Spectral Angle Mapper (SAM) algorithm is used widely in hyperspectral data processing, such as classification, detection, identification, etc. In many cases, however, the classification result of SAM is not satisfied. The aim of this study is to improve the classification precision of the Spectral Angle Mapper (SAM) algorithm through investigating the change of similarity between the reference spectra and the selected spectra, evaluated by SAM, in the feature space. The properties of result calculated by SAM algorithm are exploited in the feature space whose dimensionality is equal to the number of bands. A new method, which represses the impact caused by the additive factor in the feature space, is proposed in this paper for its improvement on performance versus traditional SAM algorithm. The spectral discriminability of the new algorithm is greatly improved by reducing the additive factor in the feature space appropriately. In order to demonstrate its enhancement, a comparative study is conducted between the new algorithm and the SAM. The comparative results prove that the new approach can control the errors effectively and improve the precision and reliability of classification significantly. The new algorithm is implemented in IDL7.0 and tested in ENVI, using 1995 Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data from Cuprite, Nevada, USA.