Comparison of spectral-spatial classification for urban hyperspectral imagery with high resolution

For urban hyperspectral imagery with high spatial resolution, both spectral and spatial information are important and should be combined together to improve classification accuracy. In this paper, different combination strategies are investigated. In particular, a two-stage algorithm is developed where the pixel shape index (PSI)-based features are extracted as low level spatial features which are combined with dimensionality-reduced spectral features as inputs to a support vector machine (SVM) for classification. Then the resulting classification is refined with high level class spatial neighborhood information to further improve the classification accuracy. The preliminary result shows the effectiveness of this two-stage algorithm.