摘要
噪声压制是地震勘探中一个长期存在的问题,虽然一些传统方法能够压制数据中的噪声,但存在有效信号丢失、噪声残留等问题。为此,提出了一种基于卷积降噪自编码器的无监督地震数据去噪算法。该算法首先对地震数据进行一定程度的随机损坏,然后将损坏后的地震数据输送到编、解码框架。编码框架负责捕捉地震数据波形特征,据此消除噪声;解码框架能够对特征图进行扩大并恢复地震数据细节信息,从而得到重构的地震数据。最后,将重构地震数据与原始地震数据之间的误差作为收敛代价进行模型训练。考虑到地震数据的复杂性与特殊性,在编码和解码阶段使用了多尺度卷积模块提取地震数据特征。合成数据与实际数据的验算结果表明,该方法在保护地震信号的同时能够有效压制随机噪声、提高地震信号的信噪比。
Noise attenuation is a long-standing problem in seismic exploration.Traditional denoising methods can suppress seismic noise,but they may cause lost effective signals,residual noises and other problems.An unsupervised denoising algorithm based on a convolutional denoising autoencoder is proposed,which can significantly improve the signal-to-noise ratio of seismic data.The algorithm locally and randomly damages seismic data,and then transmits the seismic data damaged to coding and decoding frameworks.The coding framework captures the waveform of the seismic data and eliminates the noise.The decoding framework expands the feature map,recovers the details of the seismic data.Finally,after reconstructing the seismic data,the algorithm trains a model with the error between the reconstructed seismic data and the original seismic data.Considering the complexity of seismic data,a multi-scale convolution module is required to extract the characteristics of seismic data during coding and decoding.Applications to synthetic and real seismic data have proved that the new method is more effective in preserving signals while suppressing noises.Its denoising result is better than a traditional algorithm.
作者
宋辉
高洋
陈伟
张翔
SONG Hui;GAO Yang;CHEN Wei;ZHANG Xiang(Key Laboratory of Exploration Technologies for Oil and Gas Resources(Yangtze University),Ministryof Education,Wuhan,Hubei 430100,China;College of Geophysics and Petroleum Resources,Yangtze University,Wuhan,Hubei 430100,China;CNPC Key Laboratory of Geophysical Prospecting,China University of Petroleum(Beijing),Beijing102249,China;Hubei Cooperative Innovation Center of UnconventionalOil and Gas,Wuhan,Hubei 430100,China)
出处
《石油地球物理勘探》
EI
CSCD
北大核心
2020年第6期1210-1219,1160-1161,共12页
Oil Geophysical Prospecting
基金
国家自然科学基金项目“基于经验模态分解的自由表面多次波衰减方法研究”(41804140)
“碳酸盐岩不同孔隙结构多尺度三维数字岩心建模方法研究”(41674136)联合资助
关键词
无监督学习
卷积神经网络
降噪自编码器
地震数据
去噪
unsupervised learning
convolutional neural network
denoising autoencoders
seismic data
denoising
