3-D back projection and matching for ship recognition from SAR images

A new technique for ship recognition from synthetic aperture radar (SAR) images was proposed. First, a method of projecting SAR image back to the three-dimensional (3-D) target space was developed to obtain the 3-D distribution of the observed target scatterers, namely the 3-D back projection scattering image (BPSI). Next, the 3-D hot scattering point images (HSPIs) of candidate ships as the benchmark were predicted via the physical optics (PO) method. To classify the observed ship efficiently, a two-stage matching scheme was applied: the first stage using geometric features implemented a pre-selection to reduce the number of candidate models from the classification library; the resulting models would undergo the subsequent recognition stage, where a fuzzy matching between the 3-D BPSI and 3-D HSPI was carried out easily and also less sensitive to nonideal factors such as calculation errors than point to point matching. Experimental results for simulated and real ship SAR images display the high discrimination and better visual effect of the 3-D scattering feature, and demonstrate the effectiveness of the proposed approach.