Integrating incidence-angle compensation in the segmentation of UAVSAR images

Abstract

<p>SAR images display a characteristic intensity variation along the range dimension, caused by the fact that the backscattered intensities depend on the incidence angle. In the case of wide-swath images, this intensity variation from near range to far range is significant enough to affect image segmentation performed on absolute intensity values. The effects are an over-segmentation which creates banding in the range direction, as well as the dillution of the real class distinction. In addition, the decay rates vary for different classes, thus reducing the efficiency of previously proposed global-correction methods. <p>We propose an unsupervised segmentation method that incorporates the incidence-angle variation into the standard mixture modeling. We demonstrate its efficiency on UAVSAR images containing oil spills and ships on an open-water background, acquired in the North Sea during 2015 (the NORSE2015 experiment). By considering the intensities of the HH channel and an incidence angle range spanning from 30 to 60 degrees, the proposed algorithm is able to remove the banding effect and to segment the main image structures (water, oil slicks and ships) into distinct classes, thus showing the importance of accounting for the incidence angle.