Classification of hyperspectral remote sensing data with support vector machines (SVMs) is investigated. SVMs have been introduced recently in the field of remote sensing image processing. Using the kernel method, SVMs map the data into higher dimensional space to increase the separability and then fit an optimal hyperplane to separate the data. In this paper, two kernels have been considered. The generalization capability of SVMs as well as the ability of SVMs to deal with high dimensional feature spaces have been tested in the situation of very limited training set. SVMs have been tested on real hyperspectral data. The experimental results show that SVMs used with the two kernels are appropriate for remote sensing classification problems.
