Discrimination of buried objects using angular radial transform and multi-layer perceptrons

In this study, we propose a buried object classification approach using ground penetrating radar (GPR), with special emphasis on buried surrogate mines. The processing is carried out on B-scans which are 2D GPR responses. The buried object features are extracted using angular radial transform (ART) as this method is compact, efficient and noise tolerant. Multi-layer perceptrons (MLP) are used for object classification as they can compensate for the clutter inherent in GPR responses by means of learning through examples. The classification results are compared with the output of k-nearest neighbor (k-NN) algorithm, and the superiority of neural networks is presented. The results are presented on an extensive GPR dataset consisting of several types of surrogate mines and other common objects buried under the ground.