Deep Learning-Based Time Delay Estimation Using Ground Penetrating Radar

Time delay estimation (TDE) is of great interest for the thickness estimation of pavement using ground penetrating radar (a non-destructive testing tool that uses electromagnetic waves to probe civil engineering material), which determines the difference between the times of arrival of two incoming signals or backscattered echoes. However, conventional TDE methods suffer performance degradation because of limited resolution for thin layers and highly correlated backscattered echoes. In this paper, a deep neural network (DNN)-based TDE method is proposed. Firstly, a new DNN is constructed to classify and train the backscattered echoes; then, the time delays of the backscattered echoes can be estimated through the proposed DNN. The proposed method is based on the data processing of the backscattered echoes, which is more robust to the noise than conventional subspace-based methods (MUSIC, ESPRIT) and compressive sensing-based methods (OMP). The proposed method can directly process coherent backscattered echoes without decorrelation procedures, compared with MUSIC and ESPRIT. In addition, the proposed method is more powerful in resolving the close backscattered echoes than that of OMP. Simulation results show the efficiency of the proposed method in terms of signal-to-noise ratio (SNR) and B?t products. (The B?t products indicate the resolution of GPR, B is the frequency bandwidth of GPR and ?t is the time delay between two incoming signals or backscattered echoes).

Automated summary

A deep neural network (DNN)-based time delay estimation method is proposed in this paper to estimate the time delays of backscattered echoes. This method is more robust to noise and does not require decorrelation procedures for coherent backscattered echoes, compared to conventional subspace-based and compressive sensing-based methods. Simulation results demonstrate the efficiency of this method in terms of signal-to-noise ratio (SNR) and GPR resolution.