Iceberg Detection Using Compact Polarimetric Synthetic Aperture Radar

• Main Author: Denbina, Michael
• Other Authors: Collins, Michael
• Format: Doctoral or Postdoctoral Thesis
• Language: English
• Published: Graduate Studies 2014
• Subjects: Remote Sensing; Engineering > Electronics and Electrical; Engineering > Marine and Ocean; icebergs; iceberg detection; Synthetic Aperture Radar; Circular Polarization; Compact polarimetry; Arctic; polarization; Stokes vector; Iceberg*; Sea ice
• Online Access: http://hdl.handle.net/11023/1946; https://doi.org/10.11575/PRISM/26410

The management and monitoring of icebergs and sea ice is an important consideration for any maritime operations that take place in the polar and near-polar regions. Spaceborne synthetic aperture radar (SAR) satellites can provide iceberg detection capabilities over vast areas of ocean with a quick revisit time. Canada's next Radarsat satellite, the Radarsat Constellation, will collect compact polarimetry, a polarimetric SAR configuration that transmits a circular polarization rather than the typical linear horizontal or vertical polarizations. In doing so, it can provide greater polarimetric information than typical linear dual-polarized SAR, without the swath width restrictions of fully polarimetric or quad-polarized SAR. In this thesis, the use of compact polarimetry for iceberg detection is explored. The detection performance of Radarsat Constellation data, simulated using Radarsat-2 data, is assessed using validated iceberg locations collected by a survey vessel. Many new algorithms are tested for use with compact polarimetric SAR data, including detection methods that use pseudo quad-pol reconstruction, wherein quad-pol covariance matrix elements are approximated using the compact polarimetric covariance matrix. Detection using the Stokes parameters is also tested, both through the use of the skew-normal distribution to model the Stokes vector ocean clutter, and also by using the Stokes parameters to calculate the orientation and ellipticity of the polarization ellipse. A method is proposed where the polarization ellipse parameters are then used to vary the detection threshold for each pixel. Discrimination of ships and icebergs after target detection, using a support vector machine classifier, is also demonstrated. Overall, the compact polarimetric data significantly outperformed the linear dual-polarized data, both in iceberg detection and ship/iceberg discrimination. Compact polarimetry shows much promise for maritime surveillance applications.