Incorporating Incidence Angle Variation into Sar Image Segmentation

Abstract

We present a new approach for incorporating incidence angle derived synthetic aperture radar (SAR) brightness variation directly into SAR image analysis. This approach is unique in that the incidence angle dependency is modeled explicitly into the probability density function rather than an image-wide pre-processing `correction'. It can then be used for supervised and unsupervised image analysis, and is notably able to account for a different dependency rate for each class. This has potential benefits for wide-swath SAR imagery over flat areas and ocean, wide angled airborne and UAV based SAR data, connecting narrow-beam SAR images at different acquisition angles, as well as land-based analysis with local topographic terrain angles. An initial example demonstrates unsupervised image segmentation applied to sea ice mapping for meteorological services and climate science, and is compared to the same algorithm without the incidence angle modeling.