The spectral information, characterized by land cover and land use (LCLU) classes in a
classified thematic map, is regarded as one of the most important information for remote sensing
image interpretation. The a priori knowledge about the study region, e.g. a state-wide LCLU
map, allows us to identify appropriate training sites and use a supervised classification approach
based on support vector machines (SVMs) rather than an unsupervised classification technique
like clustering or mathematical morphology [16] – [18]. The SVM is a novel type of learning
machine based on statistical learning theory introduced by Cortes, Vapnik and Burges [17], [18].
Study has been conducted addressing the SVM-based classification scheme for land cover using
polarimetric synthetic aperture radar (SAR) images [19]. In [20], it has been concluded using
hyperspectal AVIRIS images that the SVM outperforms the other traditional classification rules.
A SVM-type classifier has been developed for automatic classification of cloud data from GOES
imagery in [21] and other kernel methods for unsupervised discovery of snow, ice, clouds have
been discussed in [22]