Seismic Receiver Functions Auto-picking Method Based on Graph Convolutional Networks

Seismic receiver function is a power tool to study the underground S-wave velocity structure and the layer interface structure in seismological research. Nevertheless, valid information extraction from massive observation data requires extensive manpower and tedious data processing. Therefore, it's significant to develop an Artificial Intelligence (AI) based method to achieve the goal of small labor costs in picking of receiver functions. A new seismic receiver functions auto-picking workflow is thus presented in this work using Pearson correlation coefficient and graph convolutional networks (GCN). A first step in constructing the topological graph by means of Pearson correlation coefficient to realize preliminary selection, and then GCN method is introduced to further selection. Experimental result from MDJ station shows that model with high accuracy and recall rate can be trained on a small amount of labeled data. This method facilitates to filter out redundant invalid data and provides high-quality receiver functions for subsequent seismic analysis.