Deep recurrent neural network for ground-penetrating radar signal denoising

Abstract

The ground-penetrating radar signal is a non-linear, non-smooth signal; the detection process is susceptible to the influence of noise, so the ground-penetrating radar detection capability is reduced. In order to eliminate noise in groundpenetrating radar signals, GPR signal denoising network based on deep recurrent neural networks is proposed in the paper. We use a deep learning approach to use ground-penetrating radar signals as training data and add Gaussian noise during model training so that the network continuously learns the features of GPR signals and noise, and use GPR noise signals on the test set to verify the denoising effect of the network. Experiments demonstrate that recurrent neural networks can significantly improve the signal-tonoise ratio of noisy signals and maintain the original waveform of ground-penetrating radar signals.