摘要
针对目前地震数据随机噪声压制方法采用的稀疏表示方法中单一的正交基无法根据地震数据特征自适应调整基函数,基于数据分块的自适应超完备学习字典方法通常忽略块之间相似性的问题,提出基于多道相似组稀疏表示模型的去噪算法。由于地震记录波形在邻近记录道存在较强的相似性,首先在训练窗口内计算与目标地震数据块所包含的多道记录的波形相似度,利用相似度最高的一组数据块构造多道相似组;然后采用自适应超完备字典学习算法完成基于多道相似组的字典构建与稀疏编码;最后通过迭代阈值收缩算法求解L1范数最小优化问题,逐步提高编码系数的稀疏程度,保留地震数据主要特征,压制随机噪声。与现有随机噪声压制算法对比,本文算法具有更高的峰值信噪比(PSNR),并且能更好地保持复杂区域地震数据同相轴的局部特征。
The single orthogonal transform could not adaptively adjust basis functions according to seismic data characteristics in noise suppression based on sparse representations,and the block-based over-complete learning dictionary methods usually ignores the similarities among blocks in random noise suppression.So we propose a novel denoising algorithm based on sparse representation model of multi-trace similarity group.As there is a strong waveform similarity between adjacent traces,we first construct multi-trace similarity groups,calculate the waveform similarity of these groups with the target seismic data block in a training window,and obtain the multi-trace highest similarity group.Then the adaptively learning algorithm of overcomplete dictionary based on multi-trace similarity group is adopted to obtain a learning dictionary and sparse code.Finally,L1 norm minimization problem is solved with iterative threshold shrinkage algorithm.As the sparse degree of coding coefficients is gradually promoted,main seismic data characteristics are retained and random noise is suppressed.Comparing with the existing denoising algorithms,the proposed algorithm can yield higher peak signal-to-noise ratio(PSNR),and better preserve local event features in complex areas.
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2017年第3期442-450,共9页
Oil Geophysical Prospecting
基金
国家自然科学基金项目(61502094)
大庆市指导性科技计划项目(zd-2016-009)
东北石油大学科研培育基金项目(NEPUPY-1-22)联合资助
关键词
地震数据去噪
稀疏表示
多道相似组
字典学习
迭代阈值收缩
seismic data denoising
sparse representation
multi-trace similarity group
dictionary learning
iterative threshold shrinkage