Classification of hyperspectral data using support vector machine

A new spectral-spatial classification scheme for hyperspectral images is presented. Pixel-wise Support Vector Machines classification and segmentation are performed independently, and then the results are combined, using the majority vote approach. Thus, every region from a segmentation map defines an adaptive neighborhood for all the pixels within this region. The use of several segmentation techniques is investigated: watershed, partitional clustering and hierarchical image segmentation (HSEG). The developed classification scheme substantially improves the classification accuracies and provides classification maps with more homogeneous regions, compared to pixel-wise classification. The proposed method is especially suitable for classification of images with large spatial structures, and when different classes have dissimilar spectral responses.