Determining the precise arrival time of earthquake recordings in noisy settings is crucial for effective earthquake data processing and analysis. This study introduces a novel U-Net3+-based network specially designed for accurate first arrival picking in noisy data. To reduce information loss in seismic waveform extraction, enhancements were made to the U-Net3+ decoder, featuring a residual structure for better information propagation, a convolutional layer replacing the pooling layer to preserve high-frequency seismic waveform details, and a convolutional block attention module (CBAM) integrated into the deeper layers to prioritize critical information at the onset of the seismic wave. We evaluated our model against U-Net, U-Net2+, U-Net3+, and Phase-Net using noisy earthquake recordings, and carried out comprehensive ablation studies on each module. The model’s generalizability was further validated using the “DiTing” dataset. The proposed network model demonstrated superior accuracy and robustness against noise, with picking rates exceeding 92.42%, 90.82%, and 87.81% when signal-to-noise ratios (SNRs) dropped below 20, 10, and 5 dB, respectively. This work contributes a promising approach to developing a more noise-resistant, low-cost seismic phase picking model, and offers potential significant gains for seismic monitoring techniques.

Noise reduction is a critical step in seismic data processing. A novel strong noise reduction network is proposed in this study. The network enhances the U-Net architecture with an improved inception module and coordinate attention (CA) mechanism, suppressing noise and enhancing signal clarity. These enhancements improve the network’s capability to distinguish between signal and noise in the time–frequency domain. We trained and tested our model on the STEAD dataset, which eliminated noise across various frequency bands, improved the signal-to-noise ratio (SNR) of seismic records, and reduced the waveform distortion significantly. Comparative analyses against U-Net, DeepDenoiser, and DnRDB models, using signals with SNRs ranging from −14 dB to 0 dB, demonstrated our model’s superior performance. At the same time, we demonstrated that the Inception Conv Block has a significant impact on the denoising ability of the network. Furthermore, validation using the “Di Ting” dataset and real noisy signals confirmed the model’s generalizability. These results show that the proposed model significantly outperforms the comparative methods in terms of the SNR, correlation coefficient (r), and root mean square error (RMSE), delivering higher-quality seismograms. The enhanced phase-picking accuracy underscores the potential of our approach to advance in geophysics applications.

The velocity model is one of the main factors affecting the accuracy of microseismic event localization. This paper addresses the issue of the low accuracy of microseismic event localization in tunnels and, combined with active-source technology, proposes a “source–station” velocity model. The velocity model assumes that the velocity from the source to each station is different, and it can greatly improve the accuracy of the time-difference-of-arrival algorithm. At the same time, for the case of multiple active sources, the MLKNN algorithm was selected as the velocity model selection method through comparative testing. The results of numerical simulation and laboratory tests in the tunnel showed that the average location accuracy of the “source–station” velocity model was improved compared with that of the isotropic velocity and sectional velocity models, with numerical simulation experiments improving accuracy by 79.82% and 57.05% (from 13.28 m and 6.24 m to 2.68 m), and laboratory tests in the tunnel improving accuracy by 89.26% and 76.33% (from 6.61 m and 3.00 m to 0.71 m). The results of the experiments showed that the method proposed in this paper can effectively improve the location accuracy of microseismic events in tunnels.