【作者】 张丹丹；

【导师】 林红波；

【作者基本信息】 吉林大学 ， 信号与信息处理， 2020， 硕士

【摘要】 地震勘探噪声压制是提高地震勘探质量的重要方法。地震数据的降噪效果往往受到地震噪声特性的影响,沙漠地震勘探随机噪声的特征不同于传统的高斯白噪声,噪声的能量主要集中在与有效信号重叠的低频带,很难在抑制噪声时不损失有效信号。此外,沙漠勘探噪声具有较弱的空间相关性,在某些区域表现出与有效地震信号相似的结构特征,进一步增加了辨识有效信号的难度。因此,探索适应沙漠随机噪声特性的去噪技术,从低信噪比地震勘探数据中恢复地震勘探数据复杂的结构特征,对于提高地震勘探精度具有基础性作用。非线性复扩散(NCD)是一种有效的随机噪声压制方法,该方法通过构造非线性复扩散系数,将实数域扩散拓展到复数域,使得扩散的虚部近似为平滑的二阶导,能够有效表征边缘特征,避免对噪声的敏感性。因此,NCD能够在虚部的引导下调整扩散强度,对不同的结构特征施加不同的扩散强度。复扩散滤波在地震勘探噪声压制方面表现出良好的适应性,具备压制低频有色沙漠勘探噪声的能力。然而,NCD滤波算法采用梯度算子计算梯度方向矢量,这种基于邻域信息获得的梯度算子难以表征地震勘探同相轴的复杂结构特征,尤其对于倾角较大的同相轴。在低信噪比情况下,弱相似性沙漠随机噪声与地震勘探同相轴的结构特征具有一定的相似性,进一步影响了梯度方向估计同相轴方向的精度,复扩散滤波在压制噪声的同时,导致地震同相轴的失真。本文基于同相轴结构特征,提出基于自适应结构方向复扩散(ASOCD)的沙漠地震勘探随机噪声压制方法。该方法利用非局部径向扫描计算同相轴的方向。基于估计出的同相轴方向,构建自适应方向复扩散模型,使得扩散能沿着同相轴方向进行。此外,构建了自适应阈值,不仅能够根据虚部描述的结构调整扩散强度,还利用自适应阈值更好地辨识出弱相似噪声和有效信号。仿真实验和实际地震数据处理验证了ASOCD算法的有效性。与基于邻域信息的结构表征方法相比,利用基于非局部信息的径向扫描能够更加准确的估计沙漠地震数据中同相轴的方向,抑制噪声对方向估计的影响。基于局部协方差矩阵的自适应阈值能够很好的调节扩散的强度,从而在保留有效信号的同时加强对噪声的抑制。为了缓解大倾角同相轴失真问题,本文提出了基于张量复扩散(TCD)的沙漠地震勘探随机噪声压制模型。本文结合方向结构张量(DST)和复扩散系数构建TCD模型,利用DST的特征向量获得大倾角同相轴方向的准确估计,进而实现扩散沿着复杂结构的同相轴。同时,联合虚部和结构自适应阈值将传统的实数域扩散张量扩展到复数域,在弱相似噪声区域施加强扩散,增强抑制弱相似性噪声能力;在有效信号区域减小扩散强度,从而有效地保留边缘。相比于传统的结构张量,DST利用结构张量的特征向量空间计算同相轴的方向,则能够更加有效描述复杂结构的同相轴。仿真实验与实际地震数据处理结果验证了TCD算法能够适应复杂结构特征的地震勘探数据,与经典的实数域各向异性扩散方法相比,能够在保留信号的同时,更有效地抑制弱相似沙漠背景噪声。更多还原

【Abstract】 Seismic noise suppression is an important method to improve the quality of seismic data.The effect of seismic noise reduction is often affected by the characteristics of seismic noise,and the characteristics of desert random noise are far from the white Gaussian noise.The energy of desert random noise is mainly concentrated in the low frequency band and overlaps with the effective signals.Therefore,it is difficult to suppress the noise without losing the effective signal.In addition,desert random noise is weakly spatially correlated,and have the similar structural features with the effective seismic signals in some areas,further increasing the difficulty in identifying effective signals.Therefore,developing the denoising technology adapted to remove the desert random noise from low signal-to-noise-ratio seismic data has a fundamental role in improving the accuracy of seismic exploration.Nonlinear complex diffusion(NCD)is an effective method for suppressing random noise.This method expands the real domain to the complex domain by constructing a nonlinear complex-value diffusion coefficient,so that imaginary part can be approximated as the smoothed second derivative.The imaginary part can characterize the edge structure and avoid the sensitivity to noise.Therefore,the NCD can adjust the diffusion intensity under the guidance of the imaginary part,and apply different diffusion strengths to different structural features.Complex diffusion filtering shows good adaptability in suppressing seismic noise,and has the ability to suppress low-frequency colored desert random noise.However,the NCD filter algorithm uses a gradient operator to calculate the gradient direction.This gradient operator based on neighborhood information is difficult to characterize the complex structural characteristics of the seismic reflection events,especially for the reflection events with large slopes.In the case of low signal-to-noise ratio,the structural characteristics of the weakly similar desert random noise and the seismic reflection events have a certain similarity,which further affects the accuracy of the estimation of the reflection orientation.The complex diffusion filter causes distortion of the seismic reflection events while suppressing noise.Based on the structural characteristics of the reflection events,an adaptive structure-oriented complex diffusion(ASOCD)method is proposed for desert random noise suppression.This method uses the non-local radial scanning method to calculate the orientation of the reflection event.Based on the estimated the orientation of the reflection event,a direction-adaptive complex diffusion model is constructed so that diffusion can proceed along the orientation of the reflection event.In addition,an adaptive threshold is constructed,which can not only adjust the diffusion intensity according to the structure described by the imaginary part,but also use the threshold to better identify weakly similar noise and effective signals.Simulation experiments and real seismic data processing verify the effectiveness of the ASOCD algorithm.Compared with the structure characterization method based on neighborhood information,the radial scan based on non-local information can more accurately estimate the orientation of the reflection event in desert seismic data,suppress the influence of noise on the direction estimation.The adaptive threshold based on the local covariance matrix can well adjust the intensity of the diffusion,thereby enhancing the suppression of noise while preserving the effective signal.In order to alleviate the distortion of reflection events with rapidly varying slopes,a tensor complex diffusion(TCD)model is further developed for desert random noise suppression.In this paper,the TCD model is constructed by combining the directional structure tensor(DST)and the complex diffusion coefficient.The eigenvectors of DST are used to obtain an accurate estimate of the orientations of the reflection events with rapidly varying slopes and large slopes,thereby achieving the diffusion along the reflection events of complex structures.At the same time,combining the imaginary part and the structure adaptive threshold will extend the traditional real domain diffusion tensor to the complex domain,applying strong diffusion in the weakly similar noise region to enhance the ability to suppress weakly similar noise and reducing the diffusion strength in the effective signal region to effectively preserve the edges.Compared with the traditional structural tensor,DST calculates the orientation of the reflection event in the eigenvectors space of the structural tensor,which can more effectively describe the reflection events with complex structures.Simulation experiments and real seismic data processing verify that the TCD algorithm can adapt to seismic data with complex structural features.Compared with the classical real domain anisotropic diffusion method,the TCD can more effectively preserve signals while suppressing weakly similar desert background noise.更多还原