On the effect of synthetic morphological feature vectors on hyperspectral image classification performance

This paper studies the effect of synthetic feature vectors on the classification performance of hyperspectral remote sensing images. As feature vectors, it has been chosen to employ morphological attribute profiles, that have proven themselves in this field. At this early stage of our work, the relatively simple Bootstrapping algorithm has been used for synthetic feature vector generation. Based on experiments conducted on multiple hyperspectral datasets, it has been observed that synthetic feature vectors contribute considerably to classification performance in the case of limited training dataset sizes.