Item

Abstract

ISODATA cluster analysis was used to investigate cluster separability using the Jefferies-Matusita distance as a quantitative measure. In comparison to operational sensors, MOMS-02 clusters show lesser separation performance which is due to the higher spatial resolution resulting in higher spectral variation. Given a certain cluster distance, MOMS-02 data trends to assign more clusters in principle. Texture images calculated from PC1 and panchromatic data using co-occurance matrices and 2nd. order statistics were used to account for the higher spatial resolution. Cluster analysis confirms the improvement of the separability performance using texture images as auxiliary data. Landuse classes were used to verify the results to real life applications. Various texture descriptors in different directions were generated using panchromatic data. A separability analysis was applied to training areas of different landuse classes using spectral bands only and spectral bands in combination with extracted texture images. Generally, the class separability is improved by the use of texture information whereas different texture descriptors using different window sizes and directions are necessary for different landuse classes.