When recording seismic ground motion in multiple sites using independent recording stations one needs to recognize the presence of the same parts of seismic waves arriving at these stations. This problem is known in seismology as seismic phase picking or, more generally, seismic detection.
데이터셋
출처에 같은 태스크 이름으로 연결된 데이터셋입니다. 동명 태스크가 포함될 수 있으므로, 비교할 때는 평가 결과의 조건도 확인하세요.
These files are supplementary material for “Generalized Seismic Phase Detection with Deep Learning” by Ross et al. (2018), BSSA (doi.org/10.1785/0120180080). The models were trained using keras and TensorFlow, and can be used with these libraries. The training dataset contains 4.5 million seismograms evenly split between P-waves, S-waves, and pre-event noise classes. We encourage the use of this hdf5 dataset for training deep learning models, and hope that it and the model architecture in the paper can serve as a benchmark for future studies. For additional information please contact Zachary Ross (zross@caltech.edu).
INSTANCE is a data collection of more than 1.3 million seismic waveforms originating from a selection of about 54,000 earthquakes occurred since 2005 in Italy and surrounding regions and seismic noise recordings randomly extracted from event free time windows of the continuous waveforms archive. The purpose is to provide reference datasets useful to develop and test seismic data processing routines based on machine learning and deep learning frameworks. The primary source of this information is ISIDe (Italian Seismological Instrumental and Parametric Data-Base) for earthquakes and the Italian node of EIDA (http://eida.ingv.it) for seismic data. All the waveforms have been sized to a 120 s window, preprocessed and resampled at 100 Hz. For each trace we provide a large number of parameters as metadata, either derived from event information or computed from trace data. Associated metadata allow for the identification of the source, the station, the path travelled by seismic waves and assessment of the trace quality. The total size of the data collection is about 330 GB. Waveforms files are available either in counts or ground motion units in hdf5 format to facilitate fast access from commonly used machine learning frameworks.