Performance Analysis and Validation of Waterline Extraction Approaches Using Single- and Dual-Polarimetric SAR Data

Abstract

In this study, the performance of two waterline extraction approaches is analyzed using dual-polarization Cosmo-SkyMed (CSK) Synthetic Aperture Radar (SAR) data and ancillary ground truth information. The single-polarization approach is based on multiscale normalized cuts segmentation; while, the dual-polarization one exploits the inherent peculiarities of the CSK PING PONG incoherent dual-polarimetric imaging mode together with a tailored scattering model to perform land/sea discrimination. The two approaches are applied to the actual CSK SAR data collected over the coastal area of Shanghai, China. To provide a detailed and complete validation of the two approaches, we carried out several field surveys collecting in situ ancillary information including Global Positioning System (GPS) data and tidal information. Experimental results show that 1) both approaches provide satisfactory results in extracting waterline from CSK SAR data in the intertidal flat under low-to-moderate wind conditions and under a very broad range of incidence angles; 2) the accuracy of the waterline extracted by both approaches decreases in case of water within the intertidal flat; 3) the single-polarization approach is unsupervised when the land/sea contrast ratio is high. However, it needs manual supervision to correct the extracted waterline when the land/sea contrast is low or in complex areas. A typical CSK scene is processed in about 25 min; 4) the dual-polarization approach is unsupervised and very effective: a typical CSK SAR scene is processed in seconds.