L- and X-Band Synthetic Aperture Radar Polarized Images for Oil Spill and Ship Detection
Trivero, Paolo1; Loreggia, Davide2; Ananasso, Cristina3; Biamino, Walter1; Borasi, Maria1; Cavagnero, Marco1; Di Matteo, Lorenza1
1UniversitÓ del Piemonte Orientale "A. Avogadro", ITALY; 2Istituto Nazionale di Astrofisica, ITALY; 3Agenzia Spaziale Italiana, ITALY
The degradation of the marine environment caused by leakages from pipelines or by oil released by ships can be identified through remote sensing techniques. The aim of this work is to describe an automatic system, developed and set up for operational oil spill and ship detection from Synthetic Aperture Radar (SAR) images acquired in different bands. Features of L- and X-band are pointed out. Different algorithms are available for oil spill and ship detection. Starting from the previously developed and tested Oil Spill Automatic Detector (OSAD), we expanded the existing features by adding the capability to analyse not only C- but also L- and X-band images from ALOS and COSMO-SkyMed satellites. Moreover, a ship detection module has also been implemented. ALOS (Advanced Land Observing Satellite) has been operating from October 2006 to May 2011; it carried an L-band SAR able to acquire in single as well as crossed (HV or VH) polarisation modes. The best available resolution was 10 m with 70 km swath; the widest swath was over 250 km at 100 m resolution. COSMO-SkyMed (COnstellation of small Satellites for the Mediterranean basin Observation) is an operational constellation of 4 satellites operating in X-band, the first one launched in 2007 and the last one in 2010. Among a number of different acquisition modes and polarizations, the highest publicly available resolution is Spotlight-2 mode (1m/pixel over a 10 km x 10 km area, HH or VV polarization). For every analysed image a fixed procedure is applied. First the image is calibrated to get the Normalized Radar Cross Section, then the intensity is compensated in order to obtain uniform values from near range to far range. The next step is land masking, carried out by a specially designed algorithm starting from the NOAA GSHHS shoreline, in order to exclude land areas from analysis with the aim to improve oil slick and ship detection capabilities close to the coastline. After these preliminary steps, OSAD procedure is applied on the image. On the basis of selected geometric and radiometric parameters, dark areas are selected on the images as "candidate slicks"; these parameters are then compared with thresholds from a verified oil slicks database to evaluate the probability for every area to be an oil slick. A ship detection algorithm is also applied on the same image, with the aim to identify the ship responsible of the pollution episodes. A Global Threshold is first defined, marking as a possible ship each pixel with the intensity greater than the threshold. Then an Adaptive Threshold method is applied on the candidate ships, comparing each pixel intensity with the surroundings. Thresholds are usually obtained from image background statistics (CFAR - Constant False Alarm Ratio, Cell Average). Because of their relevance in the ship detection accuracy and in the false alarm statistics, azimuth ambiguities, caused by the aliasing of the Doppler phase, are taken into account by exploiting the different available polarization modes. Different methods are under study to perform the optimal automatic removal of uncertainties in the COSMO-SkyMed data we are using. All described features have been integrated in an automatic operational system, extensively tested during about two year of daily activity: this made possible an extensive comparison between performances of detection algorithms when applied to images in different bands, resolutions and acquisition modes with the aim to identify the most suitable and effective ones. We analyzed real oil slick cases, verified by in-situ surveys obtained from marine surveillance and measurement campaigns, in the framework of "PRIMI" and "Sea object detection with Cosmo/SkyMed" project financed by Italian Space Agency. Thanks to higher spatial resolutions available nowadays, coastal areas are now easier to monitor with SAR. Overall system performances are good: more than 85% of the slicks are correctly classified. In ship detection we also took advantage of higher resolutions to identify very small ships and floating objects, as well as to evaluate ship parameters such as dimensions and tonnage. These parameters have been validated against in-situ measurements.