深度学习在地震初至拾取中的应用综述

Application of Deep Learning in Earthquake First Break Picking

随着计算机硬件技术的不断提高,人工智能技术在各个行业的广泛应用,地震学界开始探索用深度学习算法处理地震初至波数据,为初至波的拾取研究提供了崭新的方向。为了解决传统地震波初至拾取方法对低信噪比资料拾取精度较低、算法鲁棒性较差等缺点,提高地震初至拾取速度和效率,减少耗费的人力以及人为拾取所产生的误差,该文系统地介绍了地震事件初至波拾取的常用传统方法,详细地阐述了深度学习领域内各种经典神经网络模型在地震初至波拾取中应用现状。通过实例对比分析卷积神经网络、深度信念网络、生成对抗神经网络和深度递归神经网络等模型在地震初至波拾取中的应用效果,讨论总结了深度学习在地震初至波拾取领域内应用存在问题和应用前景,为今后地震初至拾取研究提供新的思路。

With the continuous improvement of computer hardware technology and the wide application of artificial intelligence technology in various industries,the seismological community began to explore the application of depth learning algorithm to process seismic first break data,which provides a new direction for the research of first break picking.In order to solve the shortcomings of traditional seismic first break picking methods,such as low picking accuracy of low signal-to-noise ratio data and poor robustness of the algorithm,improve the speed and efficiency of seismic first break picking,and reduce the labor consumption and human error,we systematically introduce the common traditional methods of seismic event first break pickup and expound in detail the application status of various classical neural network models in the field of depth learning in seismic first break pickup.By examples,we compare and analyze the application effects of convolution neural network,depth belief network,generating antagonistic neural network and depth recursive neural network in seismic first break pickup.The existing problems and application prospects of depth learning in the field of seismic first break pickup are discussed and summarized,which provides a new idea for the research of seismic first break pickup in the future.