Realistic 3D simulation of GPR for landmine/IED detection including full antenna models, soil dispersion and heterogeneity

Abstract

We investigate the near-field effects and the influence of soil properties on the detection of improvised explosive devices (IEDs) by GPR. We achieve this via numerical studies of different scenarios, performed for three soils: a sand, a loess-loam and a gravel. We implement a detailed 3D model of a 400 MHz GPR antenna, metallic and non-metallic targets, frequency-dependent soil properties as well as subsurface heterogeneity into the FDTD simulation of electromagnetic wave propagation. These numerical simulations are performed using the open-source software gprMax. Our antenna model is set up by fitting synthetic GPR traces to experimental signals of the same antenna. A set of known antenna/material parameter values were implemented, but the remaining unknown parameters were found via a global optimization method. The optimization algorithm aims at minimizing the root-mean-square error (RMSE) between the synthetic GPR trace and a measured trace that was obtained in controlled settings. The resulting group of parameters for the antenna model results in a synthetic GPR trace that has a very high correlation to the measured signal. Both the dispersive and heterogeneous subsurface models are based on experimental/laboratory data. We used a network analyzer together with a coaxial transmission line to measure the complex permittivity of the soils. We implemented the dispersive behavior in the numerical models using a multi-Debye approach. The gravel was implemented in the numerical model using a proprietary code to translate a detailed grain size distribution thereof to a numerical gravel consisting of grains with corresponding grain sizes. Our simulation showed to conform well to field measurements on test sites where IED targets had been buried in these materials. The synthetic data can be used to analyze GPR performance of different scenarios and generate training data for target recognition.