Ship detection over single-look complex SAR images

Ship Detection over Single-Look Complex SAR Images M. Migliaccio, A. Gambardella and F. Nunziata Dipartimento per le Tecnologie, Università degli Studi di Napoli Parthenope Centro Direzionale, isola C4 - 80143Napoli, Italy Email: {maurizio.migliaccio, attilio.gambardella, ferdinando.nunziata}@uniparthenope.it Abstract-Synthetic Aperture Radar (SAR) ship detection is an important application in the context of environment and security monitoring. Ship detection techniques are generally based on statistically significant contrast between the ship and the local ocean background. Typically, high resolution (few tenths of meters) SAR images need to be considered. Such images are heavily affected by the presence of the speckle, and, for this reason, many ship detection algorithms employ constant false alarm rate (CFAR) algorithms. In this study, a different approach is proposed. The speckle is not mitigated but considered as a source of information. The ship is considered as a dominant scatterer responsible for a strong and coherent backscatter signal. Hence, the different behavior of the speckle statistics in presence of a dominant scatterer exploited. A new simple and very effective filtering technique, which is able to process high resolution SAR images, has been conceived an implemented. Experiments, accomplished over C-band Single Look Complex ESR 1/2 SAR images, show the effectiveness of this new approach for ship detection. I. INTRODUCTION Synthetic Aperture Radar (SAR) ship detection can provide useful information to coastal and fishery monitoring, oil spill detection and law enforcement agencies [1]. Physically, backscatter from ships is determined by several scattering mechanisms including direct reflection from areas perpendicular to the radar beam, corner reflections and multiple reflections from the ship and sea surface [2] which cause a bright spot over SAR images. Other factors which come into play are: construction material and characteristics of the radar instrument, such as angle of incidence, frequency, polarization, and resolution [1]. The presence of speckle, which is a fundamental property of SAR imagery, causes a random bright and dark inter-pixel variation over macroscopically homogeneous areas which limit the minimum vessel size that can be detected, since smaller vessels will become indistinguishable. Often, SAR imagery is multi-look processed to reduce the effects of speckle [1]. Many processes and objects are of relevance in the context of ship detection since they can cause false alarms. False alarms are most prevalent in non-homogeneous areas of the imagery [1]. For instance variations in ocean backscatter caused by variations in surface wind speed and direction or the transitions regions between different wind conditions. In general, the greater the wind speed or the higher the waves, the 978-1-4244-2268-5/08/$25.00 ©2008 IEEE greater the sea contribution to the radar return signal, and thus the weaker the contrast between the vessel and the ocean background. Moreover, many other processes may generate false alarms: oceanographic phenomena (e.g. atmospheric fronts, internal waves, current boundaries or breaking waves), outlying rocks, shoals, sea currents and coastal effects [1]. Generally, ship detection algorithms locate the ship signature by finding local differences of the normalized radar cross section (NRCS) of a ship from sea clutter by choosing an appropriate threshold which, in most of the cases, is set empirically [1]. A single detection threshold cannot be used for the whole image since the background backscatter changes substantially with SAR angle of incidence, wind speed, and sea state. Various approaches have been developed which automatically tune the threshold during the search for targets. Many of these algorithms are referred to as constant false alarm rate (CFAR) algorithms [2]-[5]. The use of CFAR based ship detection algorithms allows mitigating speckle noise without hampering the SAR spatial resolution at expense of a high computation load. Other approaches use cross-correlation of sub-apertures or explore a wavelet-based approach [6]-[7]. In this study, a different approach is proposed for single polarization SAR data. From an electromagnetic point of view, a ship can be considered as a dominant scatterer characterized by a strong and coherent backscatter signal. This important feature can be taken into account by evaluating the Rice factor (Rf) [8] which is expected to be sensitive to the presence of a dominant scatterer such as a ship [9]. Based on this rationale a new simple and very effective filtering technique, which is able to process high resolution SAR images is conceived and implemented. Experiments, accomplished over C-band Single Look Complex (SLC) ESR 1/2 SAR images show the effectiveness of this new approach. The rest of the paper is organized as follows. In section II the filter is described. In section III the experiments are presented and in section IV the conclusions are drawn. II. THE FILTER In this section the ship detection filter is described from an electromagnetic and practical point of view. Figure 1: Power image of a sub-image relevant to the acquisition of 21 January 2002, 10:01 UTC - ERS-2, SLCI, orbit 35318, frame 2763, descending pass (a); corresponding Rf image (b) and Ship detection result (c). In general, ships, in single polarization SAR imagery, appear as individual pixels or groups of pixels which are brighter compared to their surroundings. From an electromagnetic point of view, a ship can be considered as a dominant scatterer and it is characterized by a strong coherent component of its backscattered signal. This behavior is theoretically and practically confirmed by the physically based GK distribution model for high resolution, i.e. speckled, SAR images presented in [9]. Supported by this theoretical background, a new filtering technique to detect ships over high resolution SAR images is proposed. The proposed filter can be described according the following rationale: the presence of a non-negligible coherent component in the backscattered sea surface signal can be highlighted evaluating the Rf of the area under study [9]. The Rf represents the coherent-to-incoherent received power ratio of the backscattered signal [8]. Hence, the first stage consists in the evaluation of the Rf by means of a local window through the SAR power image. It moves and calculates the coherent power over a 3x3 pixels region as well as the incoherent received power over an enclosing, 50x50 pixels wide, background region. These are then compared to determine the Rf image. The successive step consists in an adaptive threshold algorithm designed to search for pixel values which are unusually high compared to those in the surrounding area in the Rf image. This is done by setting a threshold (t) which depends on the statistics of the surrounding area. Pixel values which lie above the threshold are considered high and therefore likely to correspond to a dominant scatterer. The threshold is evaluated for each enclosing background region according the empirical relationship: t = µRf + 6σRf where µRf is the background mean and σRf standard deviation. (1) is the background III. EXPERIMENTAL RESULTS In this section a set of meaningful experiments are shown and discussed to demonstrate the effectiveness of the proposed filtering technique for ship detection. Experiments are accomplished ERS 1/2 SLC, SAR VV polarized C-band images, is presented and discussed. The SAR images were acquired by the Active Microwave Instrument (AMI) sensor mounted on board of the ERS-1/2 satellites operated by the ESA. The nominal ground resolution is 20 meters in range and 4 meters in azimuth. To allow a simples results analysis, SAR images with 20x20 meters ground resolution are taken into account. The first case is relevant to the acquisition of 21 January 2002, 10:01 UTC (ERS-2, SLCI, orbit 35318, frame 2763, descending pass) off the coast of Tuscany (Tyrrhenian Sea). Fig.1a is the power image of a sub-image in which a ship, associated to an oil spill, is visually identified. In Figs.1b-c the Rf image and Ship detection one are shown, respectively. In this case the ship is perfectly detected and there are not any look-alikes. The second case considered is relevant to the acquisition of 16 July 1992, 9:56 UTC (ERS-1, SLCI, orbit 5234, frame 2871, descending pass) off the Tunisian coast. The showed SAR subimage (Fig.2a) regards a sea surface area characterized by the presence of a huge oil spill and a ship recognizable in the top Figure 1: Power image of a sub-image relevant to the acquisition of 16 July 1992, 9:56 UTC - ERS-1, SLCI, orbit 5234, frame 2871, descending pass (a); corresponding Rf image (b) and Ship detection result (c). Figure 3: Power image of a sub-image relevant to the acquisition of 26 July 1992, 9:42 UTC - ERS-1, SLCI, orbit 5377, frame 2889, descending pass (a); and Ship detection result (b). center. In Figs.2b-c the Rf image and Ship detection one are shown, respectively. Also in this case, only the ship is detected. The third case is related to the acquisition of 26 July 1992, 9:42 UTC (ERS-1, SLCI, orbit 5377, frame 2889, descending pass) off the Island of Malta. The showed SAR sub-image regards a sea surface area in which seven small ships are visually identified (see red circles in Fig.3a). This image represent a challenging test since small ships, which are typically associated to fishery activities, are difficult to detect in high resolution SAR mages, due to the speckle, and are lost in multi-looked ones. The filtering procedure, as shown in Fig.3b, allows detecting all the ships. The last case is relevant to the acquisition of 21 June 1992, 9:42 UTC (ERS-1, SLCI, orbit 4876, frame 2889, descending pass) off the Island of Malta. The showed SAR sub-image regards a sea surface area in which eight small ships are visually identified (see red circles in Fig.4a). This case is interesting both for the presence of small ships and for the low wind area (on the left side) in which are recognizable bright spikes in correspondence of the surface wind speed variation. This latter is a typical condition which may generate false Figure 4: Power image of a sub-image relevant to the acquisition of 21 June 1992, 9:42 UTC - ERS-1, SLCI, orbit 4876, frame 2889, descending pass (a); and Ship detection result (b). alarms. The filtering results are in accordance to what formerly experienced and, very important aspect, no false alarm has been detected. IV. CONCLUSIONS In this paper a new filtering procedure to detect ships over high resolution SAR images has been proposed. It is physically based on the high coherence of the backscattered signal associated to a ship. The filtering procedure has shown to be both computationally effective and able to operate speckled SAR images. Moreover, the proposed technique, is able to detect ships, although small, with good accuracy and with low false alarms. ACKNOWLEDGMENT Authors acknowledge the European Space Agency (ESA) Category 1 Program for providing the SAR data (C1P-2769). REFERENCES [1] D. J. Crisp, The State–of-the-Art in Ship Detection in Synthetic Aperture Radar Imagery. Defence Science and Technology Organisation Research Report, DSTO-RR-0272, May, 2004 [Online]. Available: http://www.dsto.defence.gov.au/corporate/reports/DSTO–RR–0272.pdf. [2] K.Eldhuset, “An Automatic Ship and Ship Wake Detection System for Spaceborne SAR Images in Coastal Regions,” IEEE Trans. Geosci. Remote Sens., vol.34, no. 4, pp.1010-1019, July 1996. [3] P.W. Vachon, J.W.M. Campbell, C.A. Bjerkelund, F.W. Dobson, and M.T. Rey, “Ship detection by the RADARSAT SAR: Validation of detection model predictions,” Can. J. of Remote Sens., vol.23, no.1, pp.48–59, 1997. [4] C.C. Wackerman, K.S. Friedman, W.G. Pichel, P. Clemente-Colón and X. Li, “Automatic Detection of Ships in RADARSAT-1 SAR Imagery,” Can. J. of Remote Sens., vol. 27, no.4, pp. 371-378, 2001. [5] M. Liao, C. Wang, Y Wang, and L. Jiang, “Using SAR Images to Detect Ships From Sea Clutter,” IEEE Geosci. Remote Sen. Lett., vol.5, no.2, pp.194-198, Apr. 2008. [6] P. Lombardo and M. Sciotti, “Segmentation-based Technique for Ship Detection in SAR Images,” IEE Proc.-Radar, Sonar Navig., vol.148. no.3, pp.147-159, June 2001. [7] M. Tello, C. López-Martínez, and J. J. Mallorqui, “A Novel Algorithm for Ship Detection in SAR Imagery Based on the Wavelet Transform,” IEEE Geosci. Remote Sen. Lett., vol.2, no.2, pp.201-206, Apr. 2005. [8] P. Beckmann and A. Spizzichino, The Scattering of Electromagnetic Waves from Rough Surfaces. Norwood, MA: Artech House, 1963. [9] M. Migliaccio, G. Ferrara, A. Gambardella, F. Nunziata, and A. Sorrentino, “Detection of Dark Areas and Strong Scatterers in Marine SLC SAR Images”, IEEE J. Oceanic Engineering, vol.32 , no.4 , pp 839-848, Oct. 2007.