基于块先验复扩散滤波消减地震勘探强低频随机噪声方法研究

As seismic exploration turns towards the deep resources and exploration targets such as small fault blocks, thin layers, and so on, complex geological structure and surface conditions increase the complexity and difficulty for seismic data processing. Thus, suppressing strong low-frequency seismic random noise and restoring weak seismic reflection signals play an important and fundamental role in further seismic data processing, such as complex exploration targets imaging and accurate inversion. This project starts with spatial and temporal characteristics of strong low-frequency seismic random noise, focusing on modeling patch prior for non-stationary low-frequency random noise. On this basis, an adaptive complex diffusion filter for low-frequency seismic random noise attenuation will be developed by combining the patch prior of random noise. In addition, we intend to incorporate similarity prior and statistic prior as constraint to compose a structure-oriented complex diffusion regularization filter and boosting denoising algorithm, which improve performance in edge and discontinuity preservation of seismic data, and improve fidelity of filtered signals in iteration by separating weak reflection signals from low-frequency random noise at low signal-to-noise ratio. In this way, complex diffusion filtering systems with consideration of patch noise prior and patch signal prior attenuates strong low-frequency random noise along seismic structures to preserve seismic signals and fine structures. The project will provide a stable method for low-frequency seismic random noise suppression and signal recovery for seismic data processing. It contributes to the promotion of fine seismic exploration for complex targets.

地震勘探逐步转向深层资源和小断块、超薄层等复杂勘探目标，增加了地震勘探数据复杂程度和处理难度，消减地震勘探强低频随机噪声、保真恢复弱反射信号和复杂结构，对复杂勘探目标成像和反演具有重要的基础作用。本课题以地震勘探强低频随机噪声的时空特性表征为切入点，重点研究非平稳低频随机噪声块先验模型的构建，进而指导自适应复扩散滤波方案，增强地震勘探低频随机噪声压制能力。基于相似先验和统计先验约束复扩散滤波，构建沿结构滤波的复扩散正则滤波，保持边缘和不连续结构特征，在低信噪比条件下分离弱有效信号和低频随机噪声，提升迭代过程中信号的保真度。本课题在复扩散滤波过程中综合考虑块噪声先验和块信号先验，形成沿地震勘探结构滤波的块先验复扩散滤波系统，从而在压制强低频随机噪声前提下保真恢复弱反射信号和复杂结构。该课题为地震勘探数据处理提供一种稳定的强低频随机噪声压制和信号恢复方法，有助于促进复杂勘探目标的精细地震勘探。

随着陆地地震勘探目标的日趋复杂，地震勘探资料信噪比低，且常伴随着大量具有复杂特性的低频随机噪声。本项目针对地震勘探的复杂目标和新要求，开展基于块先验的低频随机噪声数理表征及消减对策研究。将深度学习方法引入低频随机噪声先验表征，提出空间秩相关系数对低频随机噪声弱相似性定性定量地度量，分析沙漠、林带等不同地区随机噪声非平稳性、弱相似性的产生机理和表征方法，为指导低频随机噪声辨识和自适应压制对策提供理论支撑；提出基于空间分数阶梯度、复扩散虚部、结构张量等地震信号先验的多维度表征方法，利用信号先验和噪声先验构建自适应扩散系数，形成一套结构自适应复扩散降噪和信号增强对策，从而使扩散沿着复杂结构地震同相轴自适应演化，逐步压制低频地震勘探噪声，增强弱有效地震信号，突破降噪技术对不同区域、不同噪声特性的适应性局限；利用可学习高斯混合模型和深度神经网络表征低信噪比地震勘探数据的块先验，构建适应非平稳地震勘探数据的深度降噪模型，提出多层次先验融合的深度复扩散降噪理论模型和实现方法，建立传统降噪模型和深度降噪算法相融合的理论框架，为从有限样本学习适应实际地震勘探数据处理的智能降噪对策提供理论模型和技术支撑。本项目方法和研究结果均在消减不同参数合成数据、沙漠、林带等野外地震勘探记录中低频噪声的工作中得到验证。在极低信噪比情况下，新降噪方法可提升信噪比14.8-24.6dB，平均结构相似度达0.93-0.95。项目成果发表在国内外高水平期刊和学术会议17篇，授权发明专利4项。此项技术的研究有望对油气地震勘探资料智能处理技术的发展起到重要作用。