摘要
基于稀疏变换的地震数据插值可提供有效、可靠的波场,但为了适应不断增加的计算量和减少CPU计算时间,必须探寻更快速稳健的方法。本文提出一种基于曲波变换的快速梯度投影法并应用于地震数据重构。即构建一个光滑的L_1范数优化模型,并用梯度投影法求解该模型。由于曲波变换具有多尺度、多方向、各向异性等特性,可对曲线形状的同相轴进行稀疏表示,计算时利用曲波正交变换加快计算速度。数值实验结果表明,该方法显著快于目前主流的稀疏反演方法,实际数据的试算效果良好。
Sparse transformation based seismic data interpolation can efficiently provide reliable wavefield.However,fast and robust methods should be well developed to meet ever increasing computations and to reduce CPU time.In this paper,we propose a fast gradient projection method to restore seismic data based on curvelet transform.A smooth L1 norm optimization model is built and a gradient projection method is proposed to solve the new model.The orthogonality of curvelet transform is utilized to speed up the method.Numerical experiments reveal that the proposed method is much fast than the current most used sparse inverse methods and suitable for field data interpolation.
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2018年第2期251-256,共6页
Oil Geophysical Prospecting
基金
国家科技重大专项"海外重点盆地地球物理勘探关键技术研究与应用(2017ZX05032-003)"资助
关键词
波场插值
梯度投影方法
曲波变换
L1范数
规则化反演
wavefield interpolation
gradient projection method
curvelet transform
L1norm regularization
inverse problem
