Segmentation-based technique for ship detection in SAR images

A novel segmentation-based ship detection scheme is proposed to cope with the typical features of sea clutter. In particular, it is shown that the standard 2D-CFAR schemes applied to both low- and high-resolution SAR images do not allow adequate control of the false-alarm rate for nonhomogeneity and non-Gaussianity characteristics of backscattering from the sea. The introduction of an appropriate first segmentation stage allows standard CFAR techniques to be applied inside homogeneous areas. Moreover, the derived approximate CFAR performance against non-Gaussian clutter allows the detection threshold to be set to achieve the desired false alarm rate. The practical performance is demonstrated for both a set of low-resolution quick-look ERS-SAR images and a set of high-resolution single-look X-SAR/SIR-C images. This proved that the proposed segmentation-based scheme gives a very high ship detection capability for both sets, with a controlled number of false alarms in the presence of any structure or fluctuation of the background