Urban land cover mapping by spatial-spectral feature analysis of high resolution hyperspectral data with decision directed acyclic graph SVM

Classification and extraction of spatial and spectral features are investigated in urban areas from for high resolution hyperspectral imagery (HHR). The approach consists of two steps. First, the shape expansion and texture features were extracted by PSI and GLCM respectively; and spectral information was expressed by parts-based component feature generated by nonnegative matrix factorization or constrained energy match filter, which is based on mixed spectral. Second, two types of HHR features are classified by directed acyclic graph SVM. We evaluated the proposed approach with three kinds feature set on Pavia DAIS data, and the results show that the spectral and spatial classified in a fusion way by SVM improves both OA and kappa compared to spectral information only; and parts-based component feature with the spectral band also had good results.