Automatic ship detection and confidence estimates

The Norwegian Defence Research Establishment (FFI) has been doing research on ship detection in Synthetic Aperture Radar (SAR) imagery since the late 1980’s, and the Norwegian Armed Forces have used SAR images operationally since 1998. From the beginning and until now there has been a growing interest in operational ship detection. There are many key questions in operational ship detection, and one is confidence levels for ship detection. That question and other factors related to confidence levels are being looked at in this report. One way of making confidence estimates is to gather evidence that can support a decision whether the vessel is a real ship or a false alarm. The ship size, ship-to-sea contrast, morphology and ship wake detection can be used as pieces of evidence. The incidence angle of the SAR satellite (e.g. if the satellite looks near-range or far-range) will affect the ability to detect a vessel at sea. In addition, the latest SAR satellites have the possibility to acquire more images each with different polarizations. Information about incidence angle and polarization can be used to improve the confidence estimate. The algorithms described in this report can be improved over time with more validated data available. Land based and space based Automatic Identification System (AIS) data should be used to be able to provide significant input to the algorithms for validation of ship detection in SAR imagery. FFI has developed an automatic ship detection tool, AEGIR, which detects vessels in all polarisation channels. The algorithms described are implemented in AEGIR.