Training Method for Ground Bounce Removal with Ground Penetrating Radar

Abstract

Downward looking ground penetrating radar (GPR) has been considered a viable technology for landmine detection. For such a GPR with the antennas positioned very close to the ground surface, the reflections from the ground surface, i.e., the ground bounce, are very strong and can completely dominate the weak returns from shallowly buried non-metallic mines. Hence, one of the key challenges of using GPRs for landmine detection is to remove the ground bounce as completely as possible without altering the landmine return. The ground bounce varies with surface roughness and soil conditions and leads to performance degradation for many ground bounce removal algorithms without reference data selection. In this paper we introduce NHD-based reference data selection methods for ground bounce removal. Experimental results are provided to illustrate the performance of the proposed method.