On the Detection and Discrimination of Ships and Icebergs Using Simulated Dual-Polarized RADARSAT Constellation Data

Maritime surveillance is an important application of synthetic aperture radar and is one of the objectives of the RADARSAT Constellation, the next generation of Canada's RADARSAT satellites. The RADARSAT Constellation will be able to collect both circular transmit, linear receive (CTLR) compact polarimetric data and linear dual-polarization data in a variety of wide swath imaging modes. In this article, we focus on the medium-resolution and low-resolution modes in order to assess the capability of the RADARSAT Constellation to both detect ships and icebergs in ocean imagery and then to discriminate between them after detection is performed. We first performed target detection, using a method whereby the false alarm rate was adjusted from pixel to pixel based on the properties of the received polarization ellipse. We then attempted to discriminate between the detected ship and iceberg targets using a support vector machine classifier. Measuring the discrimination accuracy using stratified 10-fold cross-validation, we found that the CTLR data had an accuracy of 99.3% for the medium-resolution imaging mode compared to 96.5% for the HH-HV dual-polarization. Further work is necessary using a wider variety of data, but RADARSAT Constellation data seem to have strong potential for ocean target detection and discrimination.