Semantic Segmentation of High-Resolution Airborne SAR Images using Tomographic Information

In this paper we propose to validate a previously developed semantic segmentation method on F-SAR high-resolution tomographic data acquired on a rural forested area. The method consists in the design of relevant features that exploit the information present in tomograms and their combination with spatial features computed on image intensity and tomograms. Our main goal is to demonstrate that these features are relevant for a variety of data and classes. In our experiments we show that features computed from single-polarization tomograms lead to better results than these obtained from fully polarimetric images for classes that exhibit vertical information. This is especially the case for urban and forested areas.