Convolutional models for buried target characterization with ground penetrating Radar

Abstract

Identification of buried antipersonnel landmines with ground penetrating radar (GPR) establishes a need for application-specific scattering models and associated estimation algorithms relating the measured scattered field to target characteristics such as size, material composition and burial depth. To this end, starting from integral representations of the scattered field, we derive frequency- and time-domain convolutional models for the GPR response of buried dielectric and metal mine-like targets, including simple analytical expressions for the target transfer function/impulse response. The main steps in the derivation are the linearization of the scattering problem by either the Born or the Physical Optics approximation, the application of a new far-field backscattering representation of the half-space Green's tensor, and the introduction of point source/receiver models for the GPR antennas and the receiver chain. Using three-dimensional finite-difference time-domain (FDTD) and measured data examples, we illustrate the validity of the convolutional models and how they can be used to characterize buried targets. For the characterization, we make use of a deconvolution algorithm, termed subset selection deconvolution, which uses our target impulse response approximations as a priori information on the form of the impulse response to be estimated. The results demonstrate the possibility to determine target size and depth from the estimated impulse response with millimeter accuracy under laboratory conditions, both of which are valuable information for landmine identification.