Ship Detection Using RADARSAT-2 Fine Quad Mode and Simulated Compact Polarimetry Data.

摘要

This Technical Memorandum presents an evaluation baseline for ship detection employing RADARSAT- 2 Fine Quad (FQ) Mode imagery and an initial investigation of the potential performance for data obtained by a compact polarimetry (CP) SAR system. A CP SAR system with circular polarization on transmission and two orthogonal linear polarizations on receive was simulated using RADARSAT-2 FQ data. Polarimetric SAR (PolSAR) ship detection algorithms were applied to both the FQ and simulated CP data. From statistical decision theory, the likelihood ratio test with Neyman-Pearson criterion was used to define a decision variable. FQ, dual polarization (including simulated CP, and horizontal polarization on transmit and horizontal and vertical polarizations on receive (HH + HV) with amplitude only), and single polarization (HH) images of known ships were considered. In this study, the smallest detected ship had a length of 18 m. The results demonstrate that a quad polarization system provides the best ship detection performance in this application as compared to single and dual polarization SAR systems. The results also demonstrate that a CP system provides better performance than a conventional dual polarization system and in turn than a single polarization system. It is important to note that CP, dual polarization and single polarization systems have a swath width that is twice as wide as that of a PolSAR having the same resolution. Detection performance was characterized by calculating receiver operating characteristics for RADARSAT-2 FQ data and simulated CP data to quantify the trade-off between the probability of missed detection and the probability of false alarm. SAR detections were validated by using automatic identification system (AIS) data. In further studies, detection algorithms based on the likelihood ratio test have been optimized and the decision variable defined using the likelihood ratio test has been input into a K-distribution constant.