USCNet: Seismic P-wave arrival time pickup based on deep learning

Abstract The accurate pickup of seismic P-waves has important application value for real-time seismic data processing such as earthquake activity monitoring and earthquake disaster warning. However, existing low signal-to-noise ratio seismic data pickup algorithms often face problems of low accuracy and poor robustness. To address these issues, we designed a seismic P-wave picking network called USCNet. This network combines the advantages of U-Net and SegNet networks. The U-Net network performs well in picking accuracy, but requires a large number of learning parameters; SegNet networks are trained with fewer learning parameters, but have lower picking accuracy. The USCNet network integrates the CBAM attention mechanism, which can deeply extract key information from seismic data, thereby significantly improving the picking accuracy. We use the Stanford Earthquake Dataset (STEAD) as the data source. The experimental results show that USCNet performs significantly better than U-Net and SegNet networks on this dataset, with significant improvements in metrics such as Mean (μ), Variance (δ), Precision (P), and Recall (R), as well as significantly improved picking performance.