应用图像语义分割网络的微地震事件识别和初至拾取方法

微地震初至精确拾取是目前微地震监测关键环节之一，传统的人工拾取方法耗时长、效率低，在实际应用中容易导致无法及时预警等问题。近年来，基于深度学习的长短期记忆网络模型(LSTM)常用于微地震初至拾取，但在低信噪比环境下拾取准确率较低，且忽视了微地震数据中相邻道初至之间的关联性。针对上述问题，将微地震初至发生前、后看作是图像二分类问题，提出了一种应用图像语义分割网络的微地震事件识别和初至自动拾取方法。应用在矿井中采集的实际微地震数据进行实验，结果表明，对包含岩石破裂、工程爆破等多类型微地震事件，该方法的识别准确率较现有的深度学习方法明显提高，平均拾取误差大幅降低，特别是低信噪比数据的平均拾取误差远小于LSTM法，因而具有良好的实际工程应用价值。

Microseismic event recognition and first break pi-cking method based on image semantic segmentation network

Precise picking of microseismic first breaks is one of the key aspects of microseismic monitoring. The traditional manual pickup method is time-consuming and inefficient, and it is likely to lead to problems such as failure to provide timely war-nings in practical applications. In recent years, the long short-term memory (LSTM) network model based on deep learning has been commonly used for microseismic first break picking, but the picking accuracy is low in a low signal-to-noise ratio (SNR) environment, and the correlation between first arrivals of adjacent traces is neglected. To address the above problems, we propose a microseismic event recognition and automatic first break pickup method based on the image semantic segmentation network (D-LinkNet), which treats microseismic first breaks as a binary image classification problem. Experiments using actual microseismic data acquired in mines are conducted, and the experimental results reveal that the recognition accuracy of the method is significantly improved compared with that of the existing deep learning methods for multiple types of microseismic events including rock rupture and engineering blasting, and the average pickup error is significantly reduced; in particular, the average pickup error of low SNR data is much smaller than that of the LSTM method. Thus, the proposed method has good value for practical engineering applications.