Speaker
Description
Identifying different seismic phase of a seismograph has been the cornerstone in earthquake processing and location. Most of the efforts of the seismic phase identification is the obligation of the analyst. Deep leaning is offering a powerful tool for analyst experience transfer by learning form the events data already processed. Then, these gained knowledge can be used for upcoming processing and extract missing small events. Most of the previous seismic waveform processing for phase detection and identification were done in time domain. In this study we suggest using frequency domain in form of spectrogram of the event. Spectrogram is used to represent the varying of the seismic signal power for different frequency bands though time. Using frequency domain allows the deep learning to extract more sensitive, robust and stable features from the signal which can lead to better classification result. Earthquakes recorded by the SCSN (Southern California Earthquake Data Center) are used as training and validation data sets which were used in previous similar studied to provide a stable benchmark. The three channels waveform were used to enrich the deep learning algorithm and increases the ability for phase type discrimination.
