Noise Reduction Method and Application of Drilling Source Seismic Data for Tunnel Rock Drilling Rig Based on Transfer Learning

Aiming at the problem that the measured signals from the borehole seismic source of tunnel rock drilling rigs are characterized by data clutter,serious noise interference and occasional collection anomalies,which make it difficult to be directly applied to the neural network model trained with the simulated data.In this paper,based on the U-Net denoising neural network,mutual correlation,Automatic Gain Control (AGC) and normalization are used as the preprocessing layer of the U-Net neural network to carry out the signal noise reduction process,which effectively improves the fitting effect of the complex mapping relationship between the noisy borehole source and the noiseless seismic recordings of rock drilling cart.Then,an adversarial transfer learning neural network based on Generative Adversarial Nets (GAN) was constructed to migrate the real seismic data from the tunnel,realizing the effective application of real data in the trained U-Net denoising neural network.Finally,relying on the actual project to evaluate the method,the results show that,using the denoising method proposed in this paper,the noise interference in the seismic record can be effectively removed,and the detection results are basically in line with the actual excavation,which preliminarily verifies the feasibility and effectiveness of the method.