A Robust Dual-Stream Microseismic Phase Picking Model on Ocean Bottom Seismometers

Accurate seismic phase picking on Ocean Bottom Seismometer data is challenged by extremely low signal-tonoise ratios. We propose CWT-EQT, an ocean microseismic phase picking model based on continuous wavelet transform, with three key innovations. First, it employs a four-channel input strategy utilizing hydrophone data to compensate for noisy horizontal components. Second, it integrates a Continuous Wavelet Transform branch to effectively separate seismic signals from ocean noise via multi-scale focusing. Third, an adaptive dynamic feature fusion mechanism is designed to automatically weigh time- and frequency-domain features. Experimental results on the OBS dataset demonstrate state-of-the-art performance, reducing P - and S-wave Mean Absolute Errors to 0.16s and 0.29s, respectively, with an outlier fraction of 0.03. CWT-EQT exhibits superior robustness and precision compared to PickBlue.