Skip to Main Content
Radar scattering from an illuminated object is often highly dependent on the target-sensor orientation. In conjunction with physics based feature extraction, the exploitation of aspect-dependent information has led to successful improvements in the detection of tactical targets in synthetic aperture radar (SAR) imagery. While prior work has attempted to design detectors by matching them to images from a training set, the generalization capability of these detectors beyond the training database can be significantly improved by using the principle of structural risk minimization. In this paper, we propose a detector based on support vector machines that explicitly incorporates this principle in its design, yielding improved detection performance. We also introduce a probabilistic feature-parsing scheme that improves the robustness of detection using features obtained from a two-dimensional matching-pursuits feature extractor. Performance is assessed by considering the detection of tactical targets concealed in foliage, using measured foliage-penetrating SAR data.