The DAS With Deep Neural Network Based on DSR-Net for Fast Earthquake Recognition

Abstract

Earthquake early warning can effectively reduce the potential earthquake damage and is extremely strict to the sensor location and identification time. Distributed fiber acoustic sensing (DAS) is a novel seismic monitoring system with existing optical communication fiber as sensors and the passive seismic transducers are opt to densely deploy in harsh earthquake prone area, saving valuable time for earthquake to reach sensors. In order to shorten the earthquake identification time and increase the accuracy of earthquake identification at the same time, a fast earthquake identification method is proposed with DAS and DSR-Net (DAS Seismic Recognition Network) deep learning network. The signal time-frequency image samples are constructed to extract signal features, and DSR-Net is used for recognition. The feasibility is verified in natural earthquake monitoring, and the recognition accuracy of the first 5s seismic data is up to 88.29%. As the duration of the earthquake increases, the recognition accuracy reaches more than 93%. This method will be an important reference for earthquake early warning and natural disaster prevention.

Keywords

• Deep learning
• distributed acoustic sensing (DAS)
• earthquake early warning
• seismic recognition