<title>Improving multisource image fusion using thematic content</title>

Abstract

This study investigates the use of thematic class correspondence in the fusion of hyperspectral data with higher spatial resolution synthetic aperture radar (SAR) data. A thematic map derived from the SAR imagery is used to introduce spatial information into the hyperspectral imagery, a spatial-spectral fusion. Because the underlying physical processes measured by the imaging systems substantially differ, classes derived from one may have partial or no relationship to classes from the other. In our approach, SAR-derived class contributions to a mixed hyperspectral pixel are weighted in the fusion process based on their correspondence with spectral classes. Unconstrained and weighted least squares solutions for the resulting linear system are described. A comparison of fusion results is presented with and without use of thematic content.