Road Extraction from Multi-frequency and Polarimetric Sar Imagery

Automatic extraction of linear topographic objects, such as, roads, rivers, or power lines, is an important topic for topographic mapping. In case of roads, an automatic extraction from Synthetic Aperture Radar (SAR) imagery is rather difficult, due to the speckle effect and the limited spatial resolution in relation to the road width. In this paper we investigate the additional use of multifrequency and polarimetric SAR data for automatic road extraction. The used approach for automatic road extraction was developed at the Technische Universität München (TUM), Germany. We carry out experiments on single and on multi-frequency imagery from Xand LBand data. The extraction results are compared with reference data, which are separated into two classes: main roads and secondary roads. To use more of the polarimetric information, road extraction from the Pauli component for odd-bounce surfaces is applied, as well. The comparisons show that the extraction results depend mainly on the visibility and the widths of the roads, as well as on the geometry of the data. Presently, for main roads quite satisfying results can be achieved in geo-referenced, multilook imagery. Regarding the extractions from slant range, single look imagery, there can be noted a strong dependency of the illumination direction of the SAR.