Accurate Ground Penetrating Radar Numerical Modeling for Automatic Detection and Recognition of Antipersonnel Landmines

Abstract

Ground penetrating radar (GPR) is a promising non-invasive technology for imaging shallowly buried low-metal or non-metallic antipersonnel (AP) landmines. However, the application of GPR to the landmine problem remains nowadays a complex scientific and technical task due to the weak echoes produced by most plastic landmines and the presence of undesirable effects from antenna coupling, system ringing and interface/soil contributions (clutter). In this context, accurate simulations, which are of great help in prediction and correct interpretation of GPR output, may become crucial for an efficient detection, clutter removal and eventual classification of the mines. <br /> This work presents a full forward model of a realistic GPR scenario which includes targets, soil, ground surface and an accurate representation and radiation characteristic analysis of the considered ultra-wideband (UWB) impulse GPR system. The modeling procedure is comprehensively described and the GPR model optimized until a good agreement between measurements and simulations is achieved. The problem is solved numerically in time and frequency domains via the Finite Element Method (FEM) and using COMSOL Multiphysics Simulation Tool. <br /> The final model is then used to perform a parametric study of the scattering signatures (one- dimensional synthetic responses) by several buried landmine-like targets and a series of configurations (depth, soil conditions, target size and shape, etc.). The extracted conclusions are summarized together with some guidelines to build a representative target signature database. <br /> Finally, this research demonstrates that accurately computed signatures can be satisfactorily employed as reference waveforms for efficient clutter suppression and enhanced landmine detection/recognition. This is done through a combined strategy consisting of an energy based detection algorithm and a cross-correlation based identification technique. The latter is implemented before conducting the detection as an additional filtering step in the form of a similarity constraint between measured and synthetic reference signals. The proposed methodology is validated using experimental data acquired in a prepared inhomogeneous test field at the Leibniz Institute for Applied Geosciences LIAG in Hannover (Germany) where diverse mine simulants were buried at different depths. In particular, the application of the combined strategy to field data yields a clear improvement in the detection sensitivity, especially for those mines which are most difficult to detect through backscattered energy considerations alone. The potential of the method for target discrimination is also evidenced and quantified via Receiver Operating Characteristic (ROC) curves.