Comparison of ship detectors for polarimetric SAR imagery

This paper examines the benefits of using fully polarimetric SAR imagery for ship detection. There are three related aspects to the work. First, RadarSat-2 quad-pol stripmap imagery is used to compare the performance of various ship detection algorithms/methods. The comparison is based on the use of ROC curves. Second, circular spotlight SAR imagery is used to examine how detector performance varies with the azimuth angle. Owing to the small number of available ship samples, this analysis has limitations. Third, again using the circular spotlight SAR imagery, the variation, with respect to azimuth angle, of the elements of the polarimetric covariance matrix of both the ship and the ocean are analysed. The results are as follows. Using the stripmap imagery, it is found that exploiting both polarimetry and spatial correlations greatly improves detection performance. Using the circular spotlight imagery, it is found that detection performance is variable with respect to azimuth angle and trends are hard to discern. Moreover, the performance of the Pauli double bounce detector is generally poor. It is conjectured that ship pitching motion degrades the polarimetric backscatter response in the circular spotlight imagery. Finally, the covariance matrix analysis confirms known trends for the ocean backscatter. The implication of these trends is that the polarimetric covariance matrix is sensitive to changes in azimuth angle and hence detectors based on them are also. The data indicate the extent to which this is the case.