Oil spill detection by means of synthetic aperture radar

In this thesis work a new approach to observe sea oil spills by means of remotely sensed SAR data and physical scattering modeling has been developed. First the mahematical problem of the oil spill detection has looked for the objective definition of the feature space and of the nature of the classification problem. The oil spill classification problem is formulated as a one-class classification problem and an approach to perform a qualitatively analysis and to objectively select among the classical features, used in SAR oil spill classification procedures, is proposed. Successively, a new a speckle model for marine Single-Look Complex SAR images has been proposed. This model allowed showing that full-resolution speckled SAR images can be effectively employed to detect dark areas and strong scatterers. The new approach is based on the generalized K model (GK) which is able to embody very different scattering cases. Finally, two studies on sea oil spills observation by means of polarimetric SAR data are accomplished. In the first approach, a CFAR filter tailored for full-polarimetric SAR data has been first applied over marine images. The polarimetric CFAR filter is able to effectively identify the dark patches which may be due to oil spills. Further, it has been studied the relevance of the polarimetric features H, α and A to assist oil spill classification. Experiments results showed that the main polarimetric feature is H. In the second approach, a filtering technique based on the standard deviation of the co-polar phase difference (σφc) has been developed and tested. Experimental results showed the usefulness of using σφc to assist oil spill detection and, in particular, for distinguishing among oil spills and biogenic look-alikes.