基于物理约束U-Net网络的地震数据低频延拓

由于震源与采集技术的影响，地震勘探数据往往缺失低频信息，严重影响后续的反演和成像处理。现有的地震数据低频延拓方法，大多建立在时域数据分布特征的基础上，容易导致频率与相位信息严重损失。为解决该问题，文中提出一种基于地震波物理参数约束的U-Net深度学习网络进行地震数据低频延拓。首先，利用理论引导数据的思想组织样本，生成大量不同特征的地震数据；然后，通过结合残差跳跃连接改进的U-Net模型学习从中高频地震数据生成低频成分的非线性映射；最后，结合地震信号的物理参数约束提升对频率、相位的恢复效果。实验证明，文中所提方法对地震数据低频恢复具有较好的效果，并在频率与相位的保持上优于同类方法，对提高后续的处理与解释精度具有较高的实用价值。

Low frequency continuation of seismic data based on physically constrained U-Net network

Due to the influence of source and acquisition technology，seismic exploration data often lack low-frequency information，which will have a great impact on the subsequent inversion and imaging processing. Most of the existing low-frequency continuation methods of seismic data are based on the distribution characteristics of time-domain data，which is easy to cause serious loss of frequency and phase information. In order to solve this problem，a U-Net depth learning network based on seismic wave physical parameter constraints is proposed to carry out low-frequency continuation of seismic data. Firstly，we use the theory to guide the idea of data to orga-nize samples and generate a large number of seismic data with different characteristics； Then，the improved U-Net model combined with residual jump connection is used to learn the nonlinear mapping of low-frequency components from medium and high-frequency seismic data； Finally，combined with the physical parameter constraints of seismic signal，the recovery effect of frequency and phase is strengthened. Experiments show that this method has a strong effect on low-frequency recovery of seismic data，and is superior to similar methods in frequency and phase maintenance. It has high practical value for improving the subsequent processing and interpretation accuracy.