基于局部保持投影和主成分分析的地震数据去噪方法

A Noise Attenuation Method for Seismic Data Based on Locality Preserving Projections and Principal Component Analysis

随着地震勘探需求的增加和地震采集技术的发展,地震数据的维度逐渐加大,传统去噪方法已不能适应高维数据空间的非线性模式,导致对地震弯曲或倾斜同相轴的处理效果不佳,即容易损伤其有效反射波信号。为此,提出一种基于局部保持投影(LPP)和主成分分析(PCA)的地震数据随机噪声衰减方法。首先,利用LPP具有较好处理高维非线性数据的优良特性,用它对每个地震采样点由其近邻点进行重构,使得在地震数据去噪过程中能够有效地保留弯曲或倾斜同相轴信息;然后,使用PCA提取出特征值较大的有效反射信号,去除特征值很小的随机噪声;最后,在人工合成地震数据上的实验结果表明:与两种经典的基于曲波变换阈值去噪和基于小波变换阈值去噪方法相比,提出的方法能很好地适用于弯曲或倾斜同相轴,可以有效地去除随机噪声,且不损伤有效信号。此外,在实际地震CMP道集和叠加剖面上进行处理也得到了很好的效果。

With the increase of the demand for seismic exploration and the development of seismic acquisition technology,the dimension of seismic data is gradually increasing and the traditional denoising methods are difficult to disposal the nonlinear model of high dimensional data space,which leads to a unsatisfactory processing result on curved and inclined events of seismic profile. At the same time,it damages to the seismic curving reflection signals and remains some noise. Therefore,a random noise attenuation method of seismic data is proposed based on the local preserving projection（LPP）and principal component analysis（PCA）. First,the method uses the reconstruction of LPP to reconstruct each sample of seismic data by its neighborhoods as LPP has good characteristics of high dimension nonlinear data processing,which can effectively preserve information of the curved and/or inclined events in the process of seismic data denoising. Then,it uses PCA decomposition to separate the effective signals with larger eigenvalues from the data containing the noise,and gets rid of non-relevant random noise with small eigenvalues. Finally,experimental results on the artificially generated data show that the proposed method can apply well to curved and inclined events compared with two classic threshold denoising methods（curvelet transform based and wavelet transform based）and effectively remove the random noise,and does not damage the effective signal. Besides,good results are obtained by the proposed method on the real seismic CMP gather and stacking profile.