Hyperspectral classification using spectral magnitude and gradient

The spectral variations caused by geometry and incident illumination may influence classification accuracy using spectral information alone. In this paper, spectral gradient derived from original spectral data was combined with spectral data for improved classification. The performance of spectral gradient in lithologic mapping was evaluated. Two classification methods, i.e. spectral angle mapper (SAM) and extended one-class support vector machine (OCSVM) were used. The results showed that joint use of spectral magnitude and gradient in hyperspectral image classification outperformed the results using the spectral magnitude data alone, and thus is an effective method for hyperspectral classification.