Ground-Penetrating Radar Random Missing Data Reconstruction Method Based on U-Net Network

Data loss is a common occurrence during ground penetrating radar (GPR) detection. Inspired by the application of deep learning in seismic data, we use deep learning to reconstruct the random missing data of GPR. In this article, first, we combine the attention mechanism and residual network with the U-Net network to reconstruct the GPR missing data. Then, in order to better reconstruct the GPR data, we use a hybrid loss function of MSE+MS_SSIM. In addition, we prove the effectiveness of the proposed network by evaluating the simulation data, and comparing with a single U-Net network can get more accurate results. Finally, the experiment also proves that the proposed hybrid loss function can better reconstruct the missing GPR data.