摘要
实际采集的地震数据不可避免地受到随机噪声的干扰,并常常伴随着数据缺失,严重降低了地震数据的信噪比与横向连续性,继而降低后续地震数据处理及反演的精度.本文基于Self2Self无监督学习框架,针对含噪非规则地震数据,设计无监督地震数据插值去噪一体化方法,并基于地震道之间的相关性优化数据采样策略,对含噪的非规则地震数据进行整道伯努利采样,构建训练数据集;为降低随机噪声对插值重建的负面影响,在损失函数中引入全变分正则化项,确保恢复地震信号具有良好的横向连续性.针对异常噪声干扰,探讨了异常噪声识别-剔除策略,结合插值去噪一体化方法可以有效提高地震数据质量.不同数值算例验证了无监督Self2Self方法在地震数据插值重建及噪声衰减中的有效性,为后续地震数据处理和解释提供良好的数据支撑.
The acquired seismic data in field cases is unavoidably affected by random noise and missing seismic traces,which can significantly decrease the Signal to Noise Ratio(SNR)and horizontal continuity of seismic data,thereby decreasing the accuracy of subsequent seismic data processing and inversion.Based on the Self2Self unsupervised learning framework,we design a simultaneous interpolation and denoising algorithm for noisy irregular seismic data.The data sampling strategy is optimized for the Self2Self algorithm based on the correlation between seismic traces,i.e.,the Bernoulli-based sampling is applied along traces of the noisy irregular data to construct a training dataset.In order to reduce the negative effects of random noise on interpolation,the total variation regularized constraint is designed considering the horizontal continuity of the recovered seismic data.For erratic noise contamination,we discuss an erratic noise identification-elimination strategy,which can be combined with simultaneous interpolation and denoising algorithm to effectively improve the quality of seismic data.Various numerical examples verify the effectiveness of the proposed Self2Self-based unsupervised seismic data interpolation and denoising in providing high-quality seismic data for subsequent seismic processing and interpretation.
作者
张蕴
杨锴
王本锋
ZHANG Yun;YANG Kai;WANG BenFeng(State Key Laboratory of Marine Geology,Tongji University,Shanghai 200092,China)
出处
《地球物理学报》
北大核心
2025年第9期3575-3587,共13页
Chinese Journal of Geophysics
基金
国家自然科学基金项目(42374134,42330805,U20B6005)
上海市科委应用数学重点项目(23JC1400502)联合资助.
