Random Subsampling and Data Preconditioning for Ground Penetrating Radars

Abstract

Ground penetrating radars (GPRs) for mine detection can profit from the many advantages that compressed sensing can offer through random subsampling in terms of hardware simplification, reduced data volume and measurement time, or imagery simplification. An intrinsic antenna-ground model is used, canceling the undesired reverberation effects and the very strong reflection from the air–soil interface, producing higher detection rates, or even unmasking shallowly buried mines. Extensive Monte Carlo simulations on real GPR measurements (800–2200 MHz) show an increase in the probability of detection, yielding globally promising exploitable results, whenever the principal component analysis technique is used a as preconditioner, as well as providing lower random subsampling bounds for frequency and spatial measurements (cross range), whether applied individually or combined.