High-quality control of receiver functions using a capsule neural network

The Red Sea-Gulf of Suez-Cairo-Alexandria Clysmic-Trend in northern Egypt is the main earthquake zone in the country, with a moderate-to-high seismic hazard and a history of significant earthquakes caused by rifting and active faulting. To improve our understanding of the tectonic and seismic processes in this area, more comprehensive imaging of the crustal structure is required. This can be achieved by increasing the number of receiver functions (RFs) recorded by the seismic stations in northern Egypt and the southeastern Mediterranean. Data handling and processing should also be automated to increase process efficiency. In this study, we developed a capsule neural network for automated selection of RFs. The model was trained on a dataset containing RFs (both selected and unselected) from five broadband stations in northern Egypt. Stations SLM, SIWA, KOT, NBNS, and NKL are located in the unstable shelf region of Egypt, where limited knowledge of the deep crustal structure is available. The proposed capsule neural network achieved an average precision of 80% on the test set. The automated selection of RFs using a capsule neural network has the potential to significantly improve the efficiency and accuracy of RF analysis, as demonstrated by the stacking test. This could lead to a better understanding of crustal structure and tectonic processes in northern Egypt and the southeastern Mediterranean.