Deep Learning-Based Network for Underground Dielectric Target Reconstruction

Abstract

Ground penetrating radar (GPR) has been widely used in underground target detection. To meet the requirements of GPR echo data interpretation and achieve accurate retrieval of the position, shape and permittivity of the dielectric targets, this paper proposes a deep neural network-based target reconstruction method using the GPR full-wave inversion (FWI). By introducing Kirchhoff migration technique, the initial focusing and imaging space alignment of GPR data are achieved, and then an underground dielectric target reconstruction network (UDTR-Net) is constructed to substitute the iterative optimization process in the traditional GPR-FWI to achieve an accurate inversion. Finally, a simulation dataset containing multiple targets is constructed for training and testing the network. The test results show that the proposed method can achieve accurate reconstruction of targets - both the reconstruction accuracy and the generalization ability are improved.