A Physically-based Approach to Observe Ships in Dual-polarized Sar Data

Ship detection using marine Synthetic Aperture Radar (SAR) images has been widely persued in recent years since it is a topic of considerable interest in several application concerning environmental monitoring, marine surveillance, fisheries, . . . . However, ship detection is a very complex problem which can hardly be optimized with conventional single-polarization SARs. Radiometric information provided by one classical polarization channel (HH, VV, or HV) is not generally sufficient to effectively detect ships, and detection methods, which are generally based on a thresholding decision over the K-distributed sea clutter, are also limited [1]. As a matter of fact, many studies have been undertaken on the use of polarimetric information for radar-based ship detection. Statistically-based approaches, e.g. Constant False Alarm Rate (CFAR) filters or convolution between different polarimetric channels, are employed to detect ships in both dualand full-polarimetric SAR data [2]. Physically based approaches, based on the Cloude-Pottier decomposition parameters or on the Cameron’s coherent target decomposition (CTD), improved to better exploit polarimetric information for ship detection purposes (SSCM), have been also developed and tested on polarimetric SAR data [3]. However, neither statisticallynor physically-based approaches are able to provide logical true-false outputs. Moreover, the above mentioned ship detection algorithms often require a high computational load. In this study target symmetry properties are employed to detect ships in dual-polarimetric SAR data. The rationale of the study lies on the fact that most natural targets, as well as sea surface, are characterized by symmetry properties which make the correlation between the likeand cross-polarized channels very low. This is not the case for man-made targets, such as ships, which are characterized by no particular symmetry property. Preliminary results, obtained by processing full-polarimetric C-band RADARSAT-2 data confirm the effectiveness of the proposed approach which, provides a logical true-false output without using any threshold, and is very suitable for segmentation purposes. Moreover, the proposed approach, being based on an algebraic operation, is also computer time effective.