A Seismic Phase Detection And Picking Method Based On Deep Learning

With the rapid development of digitalization of seismic processing and the massive construction of seismic stations,onset time picking for seismic phase has become an important source for epicenter positioning and magnitude assessment.However,seismic wave recording is a very complex type of random waveform that is often accompanied by complex noise interference.Meanwhile,the exponential increase of seismic record data puts forward higher requirements and challenges for automatic picking technology of seismic phase.In recent years,deep learning has been widely used in the field of seismic science with its excellent performance.It breaks through the limitations inherent in traditional methods and provides a new method for seismic detection.This thesis mainly studies and analyzes the problem of high-efficiency and highprecision of seismic phase detection.After a brief analysis of the existing automatic seismic phase picking method and the study of CNN and LSTM,a phase locator based on CNN and a phase picker based on LSTM are constructed.Based on the phase locator and phase picker,a new seismic phase detection method is proposed based on CNN and LSTM.The proposed method in this paper is divided into two stages: seismic phase localization and onset time picking.In the phase localization stage,the 40-second sliding window is used to decompose the seismic wave into a number of segment waves without overlap.These segment waves are classified by the phase locator and event waves can be quickly extracted from the large segment of seismic waves.In the onset time picking stage,the picking problem is treated as a three-category problem.First,the 40-second event wave is divided into 400*0.1-second frames,and each frame is classified into P-wave,an S-wave,or none-P-S-wave by using phase picker.Then,the respective signal-to-noise ratio sequences of P-wave and S-wave are constructed,and the local respective peak moments are considered as the onset time of P-wave and S-wave.The proposed method and the traditional method AR picker are compared in terms of hit rate and precision.For the P-wave detection,the hit rate of the neural network method and the AR picker method both reached a high level.For S-wave detection,the neural network method is superior to AR picker method in hit rate and accuracy.