Marine Target Detection based on Dual Channel SAR Images

Abstract:
In this study, we focus on automatic algorithms for ship detection in dual polarized satellite based Synthetic Aperture Radar (SAR) products. In the methodology applied, we process the different polarization channels separately before finally fusing the detection result. We present detection results based on ENVISAT AP data, showing that a Constant False Alarm Rate (CFAR) algorithm based on the K-distribution greatly reduces the number of false alarms for both co- and cross-polarized channels compared to a much simpler adaptive thresholding algorithm. We also show some promising preliminary results from introducing the method of log-cumulants (MoLC) as a new parameter estimator for the K-distribution. Finally, we discuss the goodness-of-fit of the K-distribution, concluding that it deviates significantly from the observed histogram at certain conditions which could lead to false detections.