Deep Convolutional Neural Network for Microseismic Signal Detection and Classification HANG ZHANG, 1,2,3 CHUNCHI MA, 1,2 VERONICA PAZZI, 3 TIANBIN LI, 1,2 and NICOLA CASAGLI 3 Abstract—Reliable automatic microseismic waveform detec- tion with high efficiency, precision, and adaptability is the basis of stability analysis of the surrounding rock mass. In this paper, a convolutional neural network (CNN)-based microseismic detection network (CNN-MDN) model was established and well trained to a high degree of accuracy using a dataset with 16,000 preprocessed waveforms. By comparison with other methods, 4000 waveforms were tested to evaluate the precision, recall, and F1-score. The results revealed that the CNN-MDN demonstrated the highest performance in microseismic detection. Moreover, the low sensi- tivity of the CNN-MDN to noise of different intensities was proved by testing on semi-synthetic data. The model also possesses good generalization ability and superior performance capability for microseismic detection under different geological structure back- grounds, and it can correctly detect the microseismic events with M w C 0.5. These preliminary results show that the CNN-MDN can be directly applied to unprocessed microseismic data and has great potential in real-time microseismic monitoring applications. Keywords: Microseismic waveform, deep learning, CNN, detection and classification. 1. Introduction With the pressures of economic development and expanding social needs, the demand for various mineral resources, energy sources, and transportation infrastructure is increasing, which drives the devel- opment and utilization of subterranean spaces (deep land). As outlined in the Chinese national 13th Five- Year Plan, strategic high-tech research and deploy- ment are proposed in technologies for deep sea, deep land, deep space, deep blue (i.e., information tech- nology), and other fields. With the emergence of large underground and tunneling projects, micro- seismic monitoring has rapidly developed as a new type of technology in disaster monitoring and early warning (Xu et al. 2011, 2018; Zhao et al. 2017; Ma et al. 2018a, b; Feng et al. 2014, 2019a, b, 2020). It can evaluate the damage and safety status of sur- rounding rock by monitoring the rupture or damage vibration inside the rock mass, and thus assess and predict the potentially dangerous areas, providing a basis for early warning and control of any potential disaster (Ma et al. 2016a, b, 2018a, b; Zhang et al. 2019). The lengthy construction period, complex construction conditions, and the characteristics of all- weather uninterrupted real-time measurement in microseismic monitoring of underground and tun- neling projects result in a large number of different types of signal acquisition and accumulation. These include microseismic signals (MS) of ‘‘interest’’ (i.e., those generated by rock movement and breakage) and other kinds of microseismic signals of ‘‘disturbances’’ (MSD) (i.e., noise from human activities). The detection of MS typically relies on staff with strong practical experience and solid seismological knowl- edge, which is a time-consuming process with low efficiency, and the accuracy cannot be easily guar- anteed. Moreover, workers lacking experience in MS analysis will produce inaccurate MS detection, thus confusing the microseismic catalogue and affecting further signal analysis. In view of these issues, various automatic microseismic detection algorithms have been devel- oped over recent years, such as the amplitude-based long and short window method, waveform autocor- relation method, cross-correlation method, and the 1 State Key Laboratory of Geohazard Prevention and Geoenvironment Protection, Chengdu University of Technology, Chengdu 610059, Sichuan, China. E-mail: hang.zhang@unifi.it; zhanghang_nn720@163.com; machunchi17@cdut.edu.cn; ltb@cdut.edu.cn 2 College of Environment and Civil Engineering, Chengdu University of Technology, Chengdu 610059, Sichuan, China. 3 Department of Earth Sciences, University of Florence, Florence, Italy. E-mail: veronica.pazzi@unifi.it; nicola.casagli@unifi.it Pure Appl. Geophys. Ó 2020 Springer Nature Switzerland AG https://doi.org/10.1007/s00024-020-02617-7 Pure and Applied Geophysics