基于曲波噪声估计的K-SVD字典学习地震资料去噪  

K-SVD Dictionary Learning Seismic Data Denoising Technique Based on Curvelet Noise Estimation

作　　者：杨熙镭 刘怀山[1,2] Yang Xilei;Liu Huaishan(Key Lab of Submarine Geoscience and Prospecting Techniques, Ministry of Education, Ocean University of China, Qingdao Shandong 266100, China;Laboratory for Marine Mineral Resources, Qingdao National Laboratory for Marine Science and Technology, Qingdao Shandong 266071, China)

机构地区：[1]中国海洋大学海底科学与探测技术教育部重点实验室,山东青岛266100 [2]海洋国家实验室海洋矿产资源评价与探测技术功能实验室,山东青岛266071

出　　处：《工程地球物理学报》2021年第4期445-452,共8页Chinese Journal of Engineering Geophysics

基　　金：国家自然科学基金(编号:91958206);国家重点研发计划(编号:2017YFC0307401)。

摘　　要：常规K-SVD字典学习方法在处理实际地震资料的过程中,往往无法得到地震随机噪声的先验信息,使得相关的误差参数无法确定,只能通过大量调参来实现最优去噪效果。基于此提出了一种基于曲波噪声估计的K-SVD字典学习地震资料去噪方法,旨在通过对地震资料进行曲波变换,选取尺度系数最大且对应方向上噪声能量最大的曲波系数,来估计随机噪声标准差,再利用K-SVD字典学习方法自适应获得超完备字典,并在重构过程中根据所得噪声标准差确定最优迭代误差参数,从而进行去噪处理。理论模型和实际地震资料的处理结果表明,该算法相较于传统的去噪方法,能在压制随机噪声的同时,最大限度地保护有效信号不被切除。In the process of processing actual seismic data,the conventional K-SVD dictionary learning method often fails to obtain the prior information of seismic random noise,which makes the relevant error parameters uncertain and can only achieve the optimal denoising effect through a large number of parameter adjustment.Based on curvelet noise estimation,this paper proposes a K-SVD dictionary learning seismic data denoising method to estimate the standard deviation of random noise by selecting the curvelet coefficient with the largest scale coefficient and the largest noise energy in the corresponding direction through the curvelet transform of seismic data.Then,the K-SVD dictionary learning method is used to adaptively obtain the over-complete dictionary,and the optimal iterative error parameters are determined according to the noise standard deviation in the reconstruction process,so as to denoise.The processing results of theoretical model and actual seismic data show that,compared with the traditional denoising methods,this algorithm can suppress random noise and protect the effective signal from being removed to the maximum extent.

关 键 词：K-SVD字典学习 地震资料去噪 曲波变换 噪声估计 

分 类 号：P631.4[天文地球—地质矿产勘探]