SAR image processing using probabilistic winner-take-all learning and artificial neural networks

This paper develops a two-stage approach for the identification of ship targets in airborne synthetic aperture radar (SAR) imagery representing open ocean scenes. The first stage of the developed approach segments the SAR image using a novel neural clustering scheme, called "probabilistic winner-take-all (PWTA)". As for the second stage, it employs a backpropagation (BP) neural network to classify ships that may be found in the segmented SAR image. Experimental results are presented. These results demonstrate that the developed two-stage ship-identification approach is successful in automatically interpreting the SAR imagery even in the presence of confusing ships and natural clutter.