基于递推策略的近偏移距数据重建  

Near-offset Data Reconstruction Based on Recursive Strategy

作　　者：凡帆 李志明[1] Fan Fan;Li Zhiming(School of Mathematics and Physics,China University of Geosciences,Wuhan Hubei 430074,China)

机构地区：[1]中国地质大学(武汉)数学与物理学院,湖北武汉430074

出　　处：《工程地球物理学报》2025年第2期296-303,共8页Chinese Journal of Engineering Geophysics

基　　金：广东省自然科学基金项目(编号:2024A1515011680);重庆市自然科学基金项目(编号:2023NSCQ-MSX0207)。

摘　　要：在海洋拖缆地震数据采集中,由于观测系统的限制,近偏移距数据常常缺失,影响后续的地震资料处理和解释。本文提出了一种基于递推策略的近偏移距数据重建方法,包括:训练数据集构造、网络训练和近偏移距数据递推重建三个阶段。在训练数据集构造阶段,从观测数据中不缺的部分获取训练数据;在网络训练阶段,采用U形网络结构(U-net),以人为缺失数据作为输入,完整数据作为标签,训练网络学习缺失数据到完整数据的映射。在近偏移距数据重建阶段,通过递推策略,将大间隔缺失的重建任务分解为多个小间隔缺失的任务,从而逐步完成数据的重建。实验结果表明:相较于传统的多通道奇异频谱分析法和自适应预测误差滤波器方法,本文方法能够更准确地重建地震数据的同相轴信息及纹理细节,并使重建结果的信噪比提升了2.5~3.5 dB。In marine towed-streamer seismic data acquisition,the limitations of the observation system often result in missing near-offset data,which adversely affects subsequent seismic data processing and interpretation.This paper presents a near-offset data reconstruction method based on recursive strategy,which includes three stages:training dataset construction,network training,and near-offset data recursive reconstruction.In the training dataset construction stage,training data are obtained from the available portions of the observed data.In the network training stage,a U-shaped network(U-net)architecture is utilized,with artificial missing data as inputs and the corresponding complete data as labels,to train the network to learn the mapping from missing data to complete data.In the near-offset data reconstruction stage,the recursive strategy breaks down the task of reconstructing data with large missing gaps into multiple smaller gap-reconstruction tasks,thus gradually completing the near-offset data reconstruction.Experimental results on the multiple datasets demonstrate that,compared with traditional methods such as multichannel singular spectrum analysis(MSSA)and adaptive prediction error filters(APEF),the proposed method achieves more accurate reconstruction of phase axis information and texture details of the seismic data,and enhances the signal-to-noise ratio of the reconstructed outcomes by 2.5 to 3.5 dB.

关 键 词：近偏移距数据 自监督 U-net 递推重建 

分 类 号：P631.4[天文地球—地质矿产勘探]