摘要
地震勘探目标区域环境的复杂多变性导致采集的地震数据存在不完整或者不规则等问题,针对这一问题,本文在压缩感知相关理论的支撑下,提出了基于超完备字典学习的缺失地震数据重构方法.首先利用K-SVD字典学习技术对地震样本数据进行训练,建立超完备字典对地震数据进行稀疏表示,然后引入高斯随机采样矩阵作为测量矩阵对地震数据进行采样;在数据重构阶段采用分段正交匹配追踪算法实现缺失地震数据的重构.通过与传统的地震数据重构方法对比,本文算法的重构效果在峰值信噪比、信噪比等指标上均优于对比算法,证明了超完备字典学习方法能更好的根据地震数据特征进行稀疏表示,从而获得较好的重构效果.
Due to the complexity and variability of the environment of the target area of seismic exploration, the seismic data collected are incomplete or irregular. To solve this problem, this paper proposed the constructed method of missing seismic data based on the over-complete dictionary learning under the support of the theory of compressive sensing. First, K-Singular Value Decomposition(K-SVD) is used to train the seismic sample data and obtain an over-complete dictionary then that is used to sparsely represent the missing seismic data. Then introduce the Gaussian random sampling matrix as the measurement matrix to sample the seismic data. In the data reconstruction phase, the Stagewise Orthogonal Matching Pursuit(StOMP) algorithm is used to realize the missing seismic data reconstruction. Compared with the traditional seismic data reconstruction method, the proposed reconstruction method is superior to the contrast method in terms of the Peak Signal-to-Noise Ratio(PSNR), Signal-to-Noise Ratio(SNR) and other indicators, which proves that the over-complete dictionary learning method can better sparsely represent the seismic data characteristics and thus get better reconstruction effect.
作者
俞国庆
贾瑞生
孙圆圆
侯文龙
YU Guo-qing;JIA Rui-sheng;SUN Yuan-yuan;HOU Wen-long(College of Computer Science and Engineering,Shandong University of Science and Technology,Shandong Qingdao 266590,China;Shandong Province Key Laboratory of Wisdom Mine Information Technology,Shandong University of Science and Technology,Shandong Qingdao 266590,China)
出处
《地球物理学进展》
CSCD
北大核心
2019年第1期229-235,共7页
Progress in Geophysics
基金
国家重点研发计划课题(2016YFC0801406)
山东省重点研发计划(2016GSF120012)
中国博士后科学基金(2015M582117)
青岛博士后研究人员应用研究项目联合资助
关键词
地震数据重构
压缩感知
超完备字典学习
分段正交匹配追踪
Seismic data reconstruction
Compressive sensing
Over-complete dictionary learning
Stagewise orthogonal matching pursuit
