Clutter suppression and feature extraction for land mine detection using ground penetrating radar

In this paper, we discuss approaches for processing ground penetrating radar (GPR) data for land mine detection. We discuss two methods of clutter suppression using ARMA modeling and also examine feature extraction using both ARMA modeling and complex natural resonance (CNR) modeling. The algorithms presented have been tested on laboratory GPR data and been found to be promising techniques. In future work, the feature vectors produced by these techniques will be processed through appropriate detection algorithms.

I. Introducton and Background

The problem of detecting and removing unexploded antipersonnel (AP) land mines is a complex issue that has been the focus of intense investigation and research effort [1]–[3]. GPR is one of the more promising technologies for demining applications. GPR has in fact already been used in many other fields, from geology to civil engineering to archeology. GPR uses a wideband antenna in order to achieve better spatial resolution of the imaged objects. The data we analyze was taken by the Demining Technology Center (DeTeC) and the Free University of Brussels (VUE) with a 1 GHz antenna. The antenna provides a good compromise between resolution and penetration depth, allowing ground penetration of half a meter to a meter and resolution on the order of several centimeters, These parameters are appropriate since most AP mines-and the mines we will consider-are at least eight centimeters in diameter and are buried no deeper than ten to fifteen centimeters in the soil.