Unsupervised approach for polarimetric SAR image classification using support vector machines

In the previous works S. Fukuda et al. (2001), we developed the classification method of land cover from polarimetric SAR data using support vector machines (SVMs). As the extended study of the SVM-based classification method, an unsupervised approach is presented in this paper. Since SVM, originally a technique for pattern recognition, can not be applied to unsupervised classification straightforwardly, we propose the automatic selection scheme of representative training areas based on the number of the closest training samples to the separating hyperplane in the feature space; such samples are called the support vectors. In the experiment for a part of the AIRSAR Flevoland data including five classes of agricultural crops, the scheme performs successful classification, which can bear comparison with the supervised result.