Abstract
Automatic vessel detection based on wide-area Synthetic Aperture Radar (SAR) imagery is achieved via a statistical hypothesis test in which a ship is declared detected when the pixel intensity exceeds a pre-determined threshold. Inherent to the physical principle of SAR as a pulsed radar system, not only legitimate vessels are detected but commonly also ambiguous reflections caused by the periodic sampling of the scene. These ambiguous returns occur in both directions, azimuth and range, and can originate at land or ships. Depending on the chosen beam mode and system parameters, many also exceed the threshold and can severely degrade the overall performance quality. In the literature several elaborate technological and signal-processing techniques were proposed to either entirely avoid the emerging of ambiguities, or at least to suppress them significantly. While hardware based solutions are not readily available on existing spacecraft and potentially cost prohibitive on future ones, most processing algorithm are computationally too laborious for near real-time applications. This paper presents an alternative solution. As SAR-based ship detection is not about creating neat imagery, ambiguities may be tolerated as long as they can be reliably identified as such during a follow-on examination. This step is being proposed to be a novel additional statistical test that decides which hypothesis (vessel or ambiguity probability) was more likely to have generated the prior detected pixel-cluster. The effectiveness of the method is shown theoretically and corroborated with real RADARSAT Constellation Mission (RCM) data.

This publication has 36 references indexed in Scilit: