Preliminary Study of Matching Pursuit for Extracting GPR Rebar Reflections

摘要

Ground penetrating radar (GPR) has developed into a primary investigation tool for the non-destructive evaluation of reinforced concrete bridge deck. The collected GPR signals are generally composed of direct wave, rebar reflections, bottom reflection and noise. Before making GPR-based predictions, rebar reflections need to be extracted first. One obstacle is that the first rebar layer is in the shallow subsurface in some cases and the target signals may be overlapped with the direct waves. Another problem associated is that the testing surfaces may be rough in real field testing environments, which lead to strong uneven direct waves in the collected signals. This kind of direct wave cannot be removed by the commonly adopted average removing method. In this paper, Matching pursuits equipped with a customized dictionary are applied to deal with the above mentioned two problems. The dictionary is built based on the excitation function and the electromagnetic scattering mechanisms. Compared to the commonly used Gabor dictionary, the proposed dictionary is more adapted to the GPR wavefronts and each target signal can be reconstructed by just one atom, In addition, there is no need for a reference signal that is not always available. The performance of the proposed method is evaluated by simulation data in this paper.