Ricerc@Sapienza

We present an image-domain adaptive likelihood ratio tests (LRT) detector for low-signature target detection in forward-looking ground-penetrating radar. We exploit multiview tomographic images of the scene under investigation to iteratively adapt the statistics of the targets and clutter arising from the interface roughness. Using numerical electromagnetic data, it is shown that the proposed adaptive LRT detector provides significantly lower false-alarm rates compared with its nonadaptive counterpart while providing comparable detection performance.

The tomographic algorithms commonly used to process ground penetrating radar (GPR) data assume the scattering phenomenon activated by ideal sources. This can be considered an adequately accurate assumption if the actual GPR antenna presents almost nondirectional features, but should instead properly revisited when a target is illuminated by means of more collimated near-field distributions.