基于地震资料有效信息约束的深度网络无监督噪声压制方法

Unsupervised noise suppression method for depth network seismic data based on prior information constraint

  • 摘要: 地震资料处理是地震勘探中的关键环节,由于地下构造和地表条件的复杂性,地震资料的处理需要经过一系列复杂流程,从而形成多种不同类型的地震数据。不同种类的地震数据具有不同的数据特征,充分利用和发掘其中的数据特征,不仅可以充分发挥处理方法的技术潜力,消除各类非地质因素对地震资料处理质量的影响,同时可以增强地震资料处理的可靠性,改善地震资料的资料信噪比及分辨率,在复杂油气藏勘探开发中具有非常重要的基础作用。叠前地震成像道集(CRP)中的有效信号同相轴近似水平,叠后地震成像数据因为地层沉积的规律性,有效信号相比于随机噪声、成像画弧噪声等干扰具有规律、简单等特点。具体表现为CRP道集及叠后地震资料有效信号具有多尺度自相似性的特征,其高维Fourier (FK或FKK)域主要能量集中在低频、低波数区域。针对上述地震数据的特点,提出一种基于先验信息约束的深度网络地震资料无监督噪声压制方法。受到深度图像先验(DIP)的启发,神经网络的结构可以视为一种特殊的隐式先验信息,合理设计网络结构可以使得网络具有多尺度自相似性特征的提取能力。由于叠前地震成像道集数据和叠后地震成像数据有效信号的多尺度自相似性,而噪声不具备这一特性,因此,特定结构的网络可以从原始数据提取出有效信号,从而达到噪声压制的目的。叠前成像道集和叠后成像的实际数据随机噪声压制试验结果表明,本文方法具有良好的保真性与鲁棒性。此外,由于本文方法具有强大的特征提取能力,因此,对常规方法不易压制的弧状成像噪声也有良好的效果。

     

    Abstract: Seismic data processing is a critical step in seismic exploration. Due to the complexity of underground structure and surface conditions, seismic data processing needs to go through a series of complex processes, thus forming various types of seismic data. Different types of seismic data have different data characteristics. Exploring and making full use of the data characteristics can not only give full play to the technical potential of processing methods, eliminate the influence of various non-geological factors on the quality of seismic data processing, but also enhance the reliability of seismic data processing. Improving the signal-to-noise ratio and resolution of seismic data plays a significant role in the exploration and development of complex reservoirs. The useful signal in pre-stack seismic imaging gathers(common-reflection-point gathers) is approximately horizontal, and the useful signal in post-stack seismic imaging data is regular and straightforward compared with random noise and arc-like imaging noise because of the regularity of stratum deposition. Therefore, the corresponding FK domain is focused on low-frequency energy due to the specific characteristics of multiscale self-similarity. According to the characteristics of the above seismic data, this paper proposes an unsupervised noise suppression method for deep network seismic data based on prior information constraints. Inspired by the deep image prior (DIP), the structure of the neural network can be regarded as a kind of particular implicit prior information. The reasonable design of network structure can improve the ability of multiscale self-similarity feature extraction. Because of the multiscale self-similarity of the useful signals of pre-stack seismic imaging gather data and post-stack seismic imaging data but noise without this characteristic, the network with specific structure can extract the useful signals from the original data, so as to achieve the goal of noise suppression. The application results of pre-stack imaging gathers and post-stack imaging data show that the proposed method has good fidelity and robustness. In addition, due to its strong feature extraction ability, the proposed method also has a good effect on arc-like imaging noise not easy to suppress by conventional methods.

     

/

返回文章
返回