SPIE Startup Challenge 2015 Founding Partner - JENOPTIK Get updates from SPIE Newsroom
  • Newsroom Home
  • Astronomy
  • Biomedical Optics & Medical Imaging
  • Defense & Security
  • Electronic Imaging & Signal Processing
  • Illumination & Displays
  • Lasers & Sources
  • Micro/Nano Lithography
  • Nanotechnology
  • Optical Design & Engineering
  • Optoelectronics & Communications
  • Remote Sensing
  • Sensing & Measurement
  • Solar & Alternative Energy
  • Sign up for Newsroom E-Alerts
  • Information for:

SPIE Photonics West 2017 | Register Today

SPIE Defense + Commercial Sensing 2017 | Call for Papers

Get Down (loaded) - SPIE Journals OPEN ACCESS


Print PageEmail PageView PDF

Remote Sensing

Using synthetic aperture radar data to detect and identify ships

A novel approach shows the value of polarimetric Envisat data for a range of important security and environmental applications.
9 March 2008, SPIE Newsroom. DOI: 10.1117/2.1200802.1062

Synthetic aperture radar (SAR) systems are active sensors offering unique high spatial resolution regardless of weather or other conditions, with wide area coverage over swaths up to 500km across. Furthermore, SAR data can be acquired with great reliability to enable precision monitoring of Earth's surface. These properties make SAR data the first candidate for operational, near-real-time applications like natural crisis monitoring and surveillance of extensive areas. In the field of Earth observation, SAR is a useful tool for operational or pilot services related to security and the environment. For example, ship detection and identification based on SAR data is a key part of any service or system dealing with maritime traffic, illegal fishery, or sea border activity, or with ocean and coastal management issues such as oil spill detection and monitoring.

SAR images have high spatial resolution, so they can be used in sea surface ship monitoring. Ships are typically constructed from large flat metal sheets and hence are usually radar bright and detectable in SAR imagery.1 Oceangoing ships tend to be larger and made of metal, whereas coastal boats are smaller and may be made of wood or fiberglass. Moreover, when oil slicks occur, they are typically associated with ships.

A ship detection system generally consists of five stages, shown in Figure 1. After the image is registered, the land is masked, since most ship detectors produce false alarms when applied to land areas. In some cases, a preprocessing step is necessary to provide more accurately calibrated SAR data. The third step is the actual detection process, which uses constant false alarm rate (CFAR) or other detectors. Then a discrimination step is used to reject false alarms where target measurements or characterization of oceanographic or meteorological phenomena are available. Finally, the estimation algorithm provides information about the parameters of the detected ships for classification.

Figure 1. The steps required for a generic ship detection and identification scheme.

Methods of ship detection using SAR data are usually based either on CFAR detectors2,3 or on image transformation and, typically, the wavelet transform. Benchmarking for operational detection systems was carried out in the Detection and Classification of Maritime Traffic from Space project.4

The detection module of the system we developed5 is based on a modification of the Search for Unidentified Maritime Objects (SUMO) detector introduced by the Joint Research Center of the European Community. In our approach, the detection process is applied to small parts (1500(×)400 pixels) of a SAR image from Envisat, one of the European Earth observation satellites. The mean brightness and standard deviation are calculated for each tile, and its value is compared with three different thresholds to determine whether it can be characterized as a target. In the SUMO detector, three region-dependent factors are used, and these need to be calculated for the sea area of interest. In our work we use data from the Automatic Identification System (AIS), operating at the University of Patras, as ground truth data in order to calculate these factors for the sea region of the Patraikos Gulf, as illustrated in Figure 2.

The AIS provides information about the length, width, speed, and latitude and longitude coordinates of each ship in the area covered. Thus we can form a vector with the ground truth information for each ship, and use the vectors to test the entire ship detection system.

Figure 2. A screen from the AIS and the corresponding advanced synthetic aperture radar image from the Patraikos Gulf acquired in VH polarization.

A key stage in our team's research is the development of the feature extraction process. To form a reliable and robust feature vector, the information contained in both vertical transmitted, vertical received (VV) and vertical transmitted, horizontal received (VH) polarizations should be used. The characteristics that we used and tested were the total area of the ship, the eccentricity, the γ ratio, the pecstrum (pattern spectrum), the length of the largest ship's axis, the ratio of the image intensity in the center and edge points of the ship, and the ratio of the total intensity in VV and VH polarizations.

These features are tested in order to assess their robustness in the thresholding and detection process. The length of the ship's axis and the ratio of the image intensity have been shown to be the most effective features in different threshold levels. This is because these features describe the ship's silhouette, which remains almost invariant at different threshold levels. On the other hand, the eccentricity, pecstrum, and γ ratio are very sensitive to different threshold levels. The feature vector we use for vessel identification is

From the available scenes from the Patraikos Gulf, 92 vessels were identified and stored in our vessel database as the signature data set. Based on AIS information, the 92 ships were separated into seven categories. Classification experiments were carried out using a simplified Bayesian classifier (Euclidean distance) and the leave-one-out method. The results revealed that only 16% of the available ships were misclassified.

A prototype of the ship detection system described here is currently under development, and extensions to conventional ship detection systems are being made. A new feature of this work is the inclusion of an automated procedure for evaluating the SUMO detector factors in cases where AIS data is available. Furthermore, the proposed feature vector incorporates in compact form information from the available polarizations when it is proven to be invariant under rotation. This feature vector will be extended in the case of fully polarimetric SAR data, and the detection system will also include a decision fusion step in the case of identification in different polarizations.

Vassilis Tsagaris, Giorgos Panagopoulos, Vassilis Anastassopoulos 
Electronics Laboratory
University of Patras
Rio, Greece

Vassilis Tsagaris is a postdoctoral researcher. His research interests include remote sensing, pattern recognition, SAR data processing and classification, and data fusion. He is a reviewer for the International Journal of Remote Sensing and IEEE Transactions on Geoscience and Remote Sensing.

Giorgos Panagopoulos is currently pursuing a PhD in image processing for remote-sensing images in the Department of Physics. His main research interests include ship detection and remote sensing.

Vassilis Anastassopoulos is a professor with research interests in digital signal processing, image processing, radar signal processing, data fusion, and pattern recognition and classification. He is a member of IEEE and a reviewer for several scientific journals.