Fusion of spectral reflectance and derivative information for robust hyperspectral land cover classification

Achieving reliable ground cover maps and high classification accuracies using limited ground truth is a key challenge for hyperspectral data analysts. In this paper, we explore the benefits of combining spectral derivative information along with reflectance information for hyperspectral classification. In addition to providing useful class-specific slope information, spectral derivatives are likely to be invariant to variations in illumination and atmospheric conditions. Potential benefits of including spectral derivative information for hyperspectral classification are studied within the context of conventional dimensionality reduction and single classification systems, as well as using a recently proposed multi-classifier decision fusion system. Classification results with experimental hyperspectral data demonstrate the benefits of the proposed approach.