Classification of Unlabeled Deep Moonquakes Using Machine Learning

摘要

This paper investigates classification of deep moon-quakes. Because some waveforms in deep moonquake contain much noise and small amplitude, estimating the source using conventional means is difficult. Therefore, we use machine learning based on waveform similarity to estimate the seismic sources of moonquakes. However, when the source of moonquake is unknown, the arrival time to the observation points is not determined. Therefore, cutting the S wave of a moonquake based on the arrival time is difficult. To classify waveforms for which the arrival time is not determined, we use long waveform from the start time of event, which might contain the arrival time. Moreover, we classify 43 unlabeled moonquakes observed by Apollo 12. As a result, labels were given with high classification probability for many moonquakes.