Improved spectral-spatial classification of hyperspectral images by using spatial information in selecting markers

Recently, a new approach, based on the Hierarchical SEGmentation (HSEG), grown from automatically selected markers using Support Vector Machines (SVM), has been proposed for spectral-spatial classification of hyperspectral images. The HSEG algorithm, which combines region object finding with region object clustering, has given good performances for hyperspectral image analysis. This technique produces at its output a hierarchical set of image segmentations. This paper aims at improving this approach by using image segmentation to integrate the spatial information into the marker selection process. In this study, the markers are extracted from the classification maps obtained by both SVM and watershed segmentation algorithm. The watershed algorithm is used in parallel and independently to segment the image. It is a powerful morphological approach to image segmentation. Moreover, the class’s pixels, with the largest population in the classification map, are kept for each region of the segmentation map. Lastly, the most reliable classified pixels are chosen from among the exiting pixels as markers. Then, a marker-based HSEG algorithm is applied. Each region from the segmentation map is classified by applying a majority vote rule over the pixel-wise SVM classification results three benchmark urban hyperspectral datasets are used for our comparisons: Pavia, Berlin and DC Mall. The results of our experiment indicate that, compared to the original hierarchical approach, the marker selection using segmentation algorithm leads in more accurate classification maps. Indeed, the proposed approach achieves an approximately 4%, 6% and 5% kappa coefficient higher than the original hierarchical-based algorithm for the Pavia, Berlin, and DC Mall datasets, respectively.