Automatic arrival-time picking of P- and S-waves of microseismic events based on object detection and CNN

Abstract

Automatic, quick and accurate picking of the arrival-times of both the P- and S-waves of microseismic (MS) events is essential to real-time data processing for MS monitoring. This study proposes a two-stage arrival-time picking method that involves extracting the region of interest of the waveform and P- and S-wave arrival-time prediction. The dataset collects 6107 MS signals with low and high signal-to-noise ratios to train and test the proposed method. The YOLO network is employed to extract the MS signal region to narrow the range of arrival-time picking and significantly improve the problem of serious interference caused by excessive negative samples in the MS signal. Then, the convolutional neural network is utilised further to precisely pick the arrival-times of both the P- and S-waves. The results showed that the proposed two-stage method is more accurate than the one-stage method in picking the arrival-time. The effects of different convolutional kernel sizes and fully connected layers on the two-stage model are also discussed. Furthermore, a new multi-index comprehensive evaluation method that considers both the accuracy and the size of the model is proposed to further evaluate the proposed model's practicality. This study presents a new method suitable for both P- and S-wave arrival-time picking to enable MS monitoring, which is valuable for various rock engineering applications.