Convolutional Neural Network to Detect Deep Low-Frequency Tremors from Seismic Waveform Images

Abstract The installation of dense seismometer arrays in Japan approximately 20 years ago has led to the discovery deep low-frequency tremors, which are oscillations clearly different from ordinary earthquakes. As such tremors may be related large earthquakes, it is an important issue seismology investigate that occurred before establishing arrays. We use learning aiming detect evidence past seismic data more than 50 ago, when waveforms were printed on paper. First, we construct a convolutional neural network (CNN) based ResNet architecture extract waveform images. Experiments applying CNN synthetic images generated according seismograph paper records show trained model can correctly determine presence waveforms. In addition, gradient-weighted class activation mapping indicates tremor location each image. Thus, proposed strong potential for detecting numerous records, enable deepen understanding relations between and