Computational modeling of on-contact antennas for the detection and localization of anti-personnel landmines via ground penetrating radar

Ground-penetrating radar (GPR) is a mature technology which has developed into a popular tool for subsurface imaging; however its application in landmine detection is still in its infancy. Landmines are typically buried in dispersive soils below a rough surface where the effectiveness of conventional air-coupled GPR is limited. By utilizing ground-contact antennas the signal penetration is dramatically improved and data analysis is simplified. In order to canvas an area while achieving ground-contact with the antennas, this research proposes that the antennas be mounted to the bottom of the feet of a walking robotic platform developed by Square One Systems Design, called the Tri-Sphere Multi-Mode Mobility Platform. Using three antennas in both the transmitting and receiving modes, three unique bistatic GPR traces can be obtained from which a novel anti-personnel landmine detection and localization method is proposed. For each GPR trace, the target reflection is enhanced using circular polarization and is extracted using background removal. The full-path travel times are then determined by correlating the target reflections with a reference signal. These travel times are used to geometrically determine the target position to a single subsurface scattering point, which is identified as the potential target location. This detection method is fully autonomous, thereby allowing the robot to canvas a large amount of area and mark potential threats without any human interaction. Using a 3-dimensional finite-difference time domain model, GPR data is simulated for sixteen statistically different rough surfaces, nine different target locations, and two target casings, amounting to 288 unique simulations. The soil modeled is 10% wet Bosnian soil, which is both lossy and dispersive. For comparison, the various simulations are analyzed with both the exact simulated background response and the statistically approximated background response. Ultimately, using the approximated background, 92% of the 288 unique cases are shown to be detected, and 49% are accurately located. Additionally, when only examining targets located directly below the Tri-Sphere robotic platform, 100% of both metal and plastic landmines are detected and 97% are shown to be accurately located. Overall, this method shows potential as a strong candidate for humanitarian demining.