PRIME-DP: Pre-trained Integrated Model for Earthquake Data Processing

We propose a novel seismic wave representation model, namely PRIME-DP (Pre-trained Integrated Model for Earthquake Data Processing), specifically designed for processing seismic waveforms. Most existing models are designed to solve a singular problem. Unlike these models, PRIME-DP is capable of multi-task single station seismic waveform processing, including Pg/Sg/Pn/Sn phase picking and P polarization classification. Moreover, it can be fine-tunned to various tasks, such as event classification without architecture modifications. PRIME-DP can achieve a recall rate of over 85% for Pg and Sg phases on continuous waveforms and achieves over 80% accuracy in P polarization classification. By fine-tuning classification decoder with NeiMeng dataset, PRIME-DP achieves 95.1% accuracy on event.