摘要
针对地震数据中不同空间位置的波形变化差异较大,全局字典学习稀疏表示方法不足以最优稀疏表示复杂局部特征的问题,提出基于结构聚类字典学习稀疏表示的随机噪声压制算法。首先利用地震数据分块结构的自相似性与全局字典稀疏表示系数分布存在的规律性与冗余性,应用K-means思想对地震数据进行分块结构聚类,对每一类数据块集合采用奇异值分解(SVD)得到超完备字典,依据各个聚类中心重新编码该类地震数据块,得到原始地震数据更稀疏的表示和描述;然后建立正则化模型更新质心和地震数据估计值;最后利用双变量迭代阈值算法求解模型中双L_1范数的优化问题,得到去噪后的地震数据。对比实验表明,应用本文方法去噪后的地震数据具有较高的信噪比及较强的局部纹理保持能力,证明了算法压制随机噪声的有效性。
Since seismic waveforms vary greatly in different spatial positions,the sparse representation based on the global dictionary learning is not enough to provide optimal sparse representations of local seismic data features.Therefore we propose a random noise suppression based on sparse representations of structured-clustering dictionary learning.First the regularity and redundancy in the coefficients distribution are represented by the self similarity of seismic data block structures and the global dictionary sparse representation,the Kmeans method is used to cluster seismic data blocks.Then according to the structural features, the overcomplete dictionary is obtained by singular value decomposition (SVD)for a class of data blocks.So the seismic data blocks are recorded according to every clustering center,and the original seismic data are more sparsely represented and described.After that,the regularization model is established to update centroid and estimated values of seismic data.Finally,the dual variable iterative threshold algorithm is used to solve the optimization problem of the double L1norm in the model, and the noise components are removed.Based on our experiments,the proposed algorithm can obtain higher peak signal to noise ratio,and better local seismic textures,which proves the effectiveness of the proposed algorithm in the random noise suppression.
作者
张岩
任伟建
唐国维
Zhang Yan;Ren Weij ian;Tang Guowei(School of Computer and Information Technology,Northeast Petroleum University,Daqing,Heilongjiang 163318,China;School of Electrical Engineering&Information,Northeast Petroleum University,Daqing,Heilongjiang 163318,China;Modern Educational Technology Center,Northeast Petroleum University,Daqing,Heilongjiang 163318,China)
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2018年第6期1119-1127,I0001,共10页
Oil Geophysical Prospecting
基金
国家自然科学基金项目(61374127)
东北石油大学青年科学基金项目(2018QNL-49)联合资助
关键词
噪声压制
稀疏表示
结构聚类
字典学习
seismic data denoising
sparse representation
structured clustering
dictionary learning