Abstract:

Magnetic resonance sounding (MRS) is a geophysical method for directly detecting groundwater. It is widely used in water resources investigation and other fields, as it is quantitative, accurate, and efficient. The quality of MRS signal strongly affects the interpretation of magnetic resonance data. In this paper, the random noise suppression of MRS signal under strong environmental interference is studied. Based on the framework of convolutional neural networks (CNN), the nonlinear mapping relationship between the time spectrum of noise signal and the time spectrum of original noiseless signal is obtained by using the training method of supervised learning, so as to suppress the noise of magnetic resonance signal. The simulation results show that the signal-to-noise ratio of noisy MRS signal can be improved by more than 15 dB. The noise suppression effect of CNN is analyzed and compared with that of the time-frequency peak filtering (TFPF) method. The effectiveness and superiority of this method is proved by the noise suppression of the measured field data. 

Key words: groundwater, magnetic resonance sounding, convolutional neural networks, noise suppression