Application of Deep Learning to Seismic Event Classification in the
Gujarat Region, India
Abstract
In anticipation to substitute the existing manual/semi-automated methods
for classifying quarry blasts, earthquakes, and noise, we developed
three convolutional neural network (CNN) models. The three CNN models
extract relevant features from seismograms (waveform), spectrograms
(spectrum), and a combination of the two respectively. A total of 3414
samples were extracted from the three categories, 15% of the data from
each category were split for testing, and the remaining data were
augmented and used for training. The waveform model, spectrogram model,
and combined model achieved accuracies of 95.32%, 93.13%, and 93.96%,
respectively. The reliability of these models was ascertained by
promising accuracies of >90% and 100% obtained for large
and small datasets from testing with SCEDC data and records from the
Palitana region (Gujarat) respectively. The results of this study
demonstrate the potential of deep learning-based approaches for the
effective classification of seismic events.