[S-TT37] 最先端ベイズ統計学が拓く地震ビッグデータ解析

[STT37-P02] Convolutional Neural Network to Detect Deep Low-Frequency Tremors from Seismic Waveform Images

*金子 亮介^1,2、長尾 大道^2,1、伊藤 伸一^2,1、小原 一成²、鶴岡 弘² (1.東京大学大学院情報理工学系研究科、2.東京大学地震研究所)

キーワード：深部低周波微動、畳み込みニューラルネットワーク、ResNet、Grad-CAM

The establishment of Hi-net [1, 2] led to the discovery of deep low-frequency tremors [3]. Considering the expected relations between tremors and large earthquakes (e.g., [4]), it is important to investigate tremors that occurred before establishing the dense seismometer arrays. Past seismometers used more than 50 years ago drew waveforms continuously on paper wrapped on a drum. The digitization of seismograph paper records by tracing the waveforms is effective for investigating large earthquakes because such waveforms are distinctive and consequently extractable given the low frequencies and large amplitudes. In contrast, tremors have much smaller amplitudes and higher frequencies than large earthquakes, so their digitization is much more difficult due to overlapping waveforms. Therefore, we use a convolutional neural network (CNN) aiming to detect evidence of tremors from seismograph paper records. The CNN is a representative deep learning method that exhibits high performance in tasks such as image recognition and handwriting recognition. A CNN can automatically tune its internal parameters by learning the characteristics of tremors from input images without requiring prior knowledge of tremors or manually adjusting the parameters. Training a CNN from scratch with real data polluted by a variety of noises may hinder the model construction and hyperparameter tuning. Thus, we conducted numerical experiments to train a CNN with synthetic images generated according to seismograph paper records. We constructed the CNN based on the ResNet [5] architecture for better performance. The results show that the trained model can learn tremor features and correctly determine the presence of tremors in the seismic waveforms. In addition, heatmaps generated based on gradient-weighted class activation mapping (Grad-CAM) [6] clearly indicate the tremor location on each image. These suggest that a CNN can be a promising alternative for effective tremor detection compared to individual waveform extraction through digitization. Based on the finding from the experiments, we will conduct CNN training with real data and apply the trained model to seismograph paper records.

References
[1] Obara, K., Kasahara, K., Hori, S., Okada, Y.: A densely distributed high-sensitivity seismograph network in Japan: Hi-net by National Research Institute for Earth Science and Disaster Prevention. Review of Scientific Instruments 76(2), 021301 (2005). doi: 10.1063/1.1854197
[2] Okada, Y., Kasahara, K., Hori, S., Obara, K., Sekiguchi, S., Fujiwara, H., Yamamoto, A.: Recent progress of seismic observation networks in Japan —Hi-net, F-net, K-NET and KiK-net—. Earth, Planets and Space 56(8), xv–xxviii (2004). doi: 10.1186/BF03353076
[3] Obara, K.: Nonvolcanic deep tremor associated with subduction in southwest Japan. Science 296(5573), 1679–1681 (2002). doi: 10.1126/science.1070378
[4] Obara, K., Kato, A.: Connecting slow earthquakes to huge earthquakes. Science 353(6296), 253–257 (2016). doi: 10.1126/science.aaf1512
[5] He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 770–778 (2016). doi: 10.1109/CVPR.2016.90
[6] Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: Grad-CAM: visual explanations from deep networks via gradient-based localization. In: 2017 IEEE International Conference on Computer Vision (ICCV). pp. 618–626 (2017). doi: 10.1109/ICCV.2017.74