Ship classification in high and very high resolution satellite SAR imagery

Abstract

To serve the security of the maritime domain, ship self-reporting systems provide information on the cooperative vessels. However, non-reporting ships should be also monitored. Satellite images can be used to detect and classify non-reporting ships. Synthetic Aperture Radar (SAR) offers monitoring capabilities regardless of clouds or daylight, and hence it is used for satellite global monitoring. Different satellite SAR systems are deployed, from European ones such as Sentinel-1, to national ones such as TerraSAR-X, presenting very diverse characteristics from their coverage to their image resolution. In this paper, two ship classification methods are presented, a method developed for use on high (20 m) resolution SAR images (Sentinel-1 dataset), and a method developed for use on very high (3 m) resolution ones (TerraSAR-X dataset). In a cross-application experiment, both methods are evaluated on both datasets. The exercise quantifies the methods’ performance across resolutions, highlighting their pros and cons in this challenging application.