Abstract: In order to reduce the data dimensionality of hyperspectral image and improve recognition efficiency, a new terrain classification method, i. e. , KDLDA subspace method, was presented. Firstly, kernel direct linear discriminant a- nalysis (KDI_DA) was used to extract nonlinear discriminant features, and then shortest distance classifier was used to perform terrain classification in the KDLDA feature subspace. The solution of KDLDA under the ordinary form of class prior possibility was also deduced. Recognition results based on airborne visible八nfrarcd imaging spectrometer (AVIRIS) hyperspectral image show that, comparing with original space method, I_DA subspace method, direct linear discriminant analysis (DLDA) subspace method, and kernel linear discriminant analysis (KLDA) subspace method, the presented KDI_DA subspace method can remarkably improve recognition efficiency.

Key words: Terrain classification, Nonlinear discriminant feature, Kernel direct linear discriminant analysis (KDLDA),Hyperspectral image