Extraction of discriminant features from impulse radar data for classification of buried objects

This paper deals with the problem of detecting and classifying buried objects. The application in mind when addressing this problem is the detection of buried landmines. Modern landmines are to a large extent made out of plastic and ceramic materials. This makes detection with traditional sensors such as metal detectors and magnetometers almost impossible. Another problem with these sensors is the high false alarm rate induced by metallic debris from exploded bomb shells. A sensor type that seems to have capability to overcome these problems is the impulse radar. The impulse radar can detect nonmetallic objects buried in the ground. The large bandwidth of the radar also gives additional information that can be used for classification purposes. The classification abilities enable discrimination between mines and stones and metallic debris, thus reducing the false alarm rate. An important step towards good classification results is to extract a set of features from measured data. The present paper elaborates on properties that an admissible feature type must possess and shows that the choice of features should be related both to the type of measurements and the type of classifier used. A number of different feature types are finally evaluated using measured data from an impulse radar system