Erratic noise suppression using iterative structure‐oriented space‐varying median filtering with sparsity constraint

Erratic noise often has high amplitudes and a non‐Gaussian distribution. Least‐squares–based approaches therefore are not optimal. This can be handled better with non–least‐squares approaches, for example based on Huber norm which is computationally expensive. An alternative method has been published which involves transforming the data with erratic noise to pseudodata that have Gaussian distributed noise. It can then be attenuated using traditional least‐squares approaches. This alternative method has previously been used in combination with a curvelet transform in an iterative scheme. In this paper, we introduce a median‐filtering step in this iterative scheme. The median filter is applied following the slope direction of the seismic data to maximally preserve the energy of useful signals. The new method can suppress stronger erratic noise compared with the previous iterative method, and can better deal with random noise compared with the single‐step implementation of the median filter. We apply the proposed robust denoising algorithm to a synthetic dataset and two field data examples and demonstrate its advantages over three different noise attenuation algorithms.     

基于数据增广和CNN的地震随机噪声压制

卷积神经网络(Convolutional Neural Network,CNN)是一种基于数据驱动的学习算法,简化了传统从特征提取到分类的两阶段式处理任务,被广泛应用于计算机科学的各个领域.在标注数据不足的地震数据去噪领域,CNN的推广应用受到限制.针对这一问题,本文提出了一种基于数据生成和增广的地震数据CNN去噪框架.对于合成数据,本文对无噪地震数据添加不同方差的高斯噪声,增广后构成训练集,实现基于小样本的CNN训练.对于实际地震数据,由于无法获得真实的干净数据和噪声来生成训练样本集,本文提出一种直接从无标签实际有噪数据生成标签数据集的方法.在所提出的方法中,我们利用目前已有的去噪方法从实际地震数据中分别获得估计干净数据和估计噪声,前者与未知的干净数据具有相似纹理,后者与实际噪声具有相似的概率分布.人工合成数据和实际数据实验结果表明,相较于F-X反褶积,BM3D和自适应频域滤波算法,本文方法能更好地压制随机噪声和保护有效信号.最后,本文采用神经网络可视化方法对去噪CNN的机理进行了探索,一定程度上解释了网络每一层的学习内容.     

利用噪声研究地下介质速度结构及其变化在中国大陆的应用

背景噪声层析成像技术已被广泛应用于地壳和上地幔速度结构的研究,这种方法不依靠地震的发生和人工源爆破,只需记录连续的噪声信号而无需产生信号,因为噪声穿过地下介质时会携带信息,然后通过利用台站记录到的连续背景噪声数据进行互相关计算和叠加,即可得到台站间的经验格林函数,从而获取对地下结构的认识。这种方法已经很好地应用于中国的东北地区、华北克拉通、青藏高原以及华南地区,并成功地揭示了这些地区地壳与上地幔顶部的速度结构。此外近年来,一些学者还利用噪声互相关技术研究地下介质地震波速度随时间的变化,通过对汶川大地震前后连续噪声记录的研究发现,大震发生后呈现同震波速降低和震后波速逐渐恢复的特点,这表明可以通过观测地震波特性的变化来监测地下应力的变化,从而为大震的预测预防工作提供科学依据。本文主要综述了近些年来背景噪声技术及其在中国大陆地区的应用。     

Efficient and accurate edge‐preserving smoothing for 3D hexagonally sampled seismic data

The automatic detection of geological features such as faults and channels is a challenging problem in today's seismic exploration industry. Edge detection filters are generally applied to locate features. It is desirable to reduce noise in the data before edge detection. The application of smoothing or low‐pass filters results in noise suppression, but this causes edge blurring as well. Edge‐preserving smoothing is a technique that results in simultaneous edge preservation and noise suppression. Until now, edge‐preserving smoothing has been carried out on rectangular sampled seismic data. In this paper, an attempt has been made to detect edges by applying edge‐preserving smoothing as a pre‐processing step in the hexagonally sampled seismic‐data spatial domain. A hexagonal approach is an efficient method of sampling and has greater symmetry than a rectangular approach. Here, spiral architecture has been employed to handle the hexagonally sampled seismic data. A comparison of edge‐preserving smoothing on both rectangular and hexagonally sampled seismic data is carried out. The data used were provided by Saudi Aramco. It is shown that hexagonal processing results in well‐defined edges with fewer computations.     

基于渐进式网络处理的低剂量Micro-CT成像方法

微米级计算机断层扫描（Micro-CT）的应用领域很广泛,在生物医学、材料科学等都有研究,而且最近几年的发展很迅速。Micro-CT图像常因辐射剂量的限制而出现噪声,所以开发一种合适的算法来抑制Micro-CT图像中噪声变得很重要。Micro-CT图像的噪声水平与扫描样本、扫描参数等参数有关,合适的噪声抑制算法应该在不同噪声水平下都有不错的性能。过去Micro-CT图像的噪声抑制算法主要是迭代重建算法,但迭代重建算法速度比较慢。深度学习方法作为近些年比较热门的图像处理方法,在临床低剂量CT图像处理上相比于传统方法效果更好、处理速度更快,有进一步在低剂量Micro-CT图像处理中应用的潜力。另外,生成对抗网络在保持图像细节上有着比卷积神经网络更好的效果,本文构建普通卷积神经网络与生成对抗网络,用于对比两者的性能差异。限制于放射源的功率,低噪声的Micro-CT数据难以获取,提出一种创新的扫描方式,可有效获取低噪声的Micro-CT数据,并对实验小鼠的进行了扫描。针对低剂量Micro-CT中较高的噪声,结合Micro-CT的成像原理,提出渐进式的低剂量Micro-CT图像处理方法,分别在解析重建前后对小鼠的Micro-CT图像进行两次处理。相比于只在断层图像上处理,渐进式方法对高噪声数据的处理效果更好。从客观指标与主观视觉效果上,分析和对比了渐进式方法与其他深度学习方法或迭代重建算法的差别。定量分析不同噪声水平下的Micro-CT图像的处理效果,分析生成对抗网络在渐进式Micro-CT图像处理中的优势与限制。渐进式Micro-CT图像处理方法生成的图像质量高、生成速度快、鲁棒性高、客观指标高,可以对进一步的高级应用如图像分割、图像三维可视化等提供帮助。      

Enhancement of double integration procedure through spectral subtraction

The presence of noise during dynamic measurements is one of the main sources of errors during integration procedures. To overcome this problem a methodology involving spectral subtraction of the noise amplitude from the signal is presented where noise characteristics are specified during ‘silence’ periods. This technique has been used for speech enhancement, and in this paper it is used to improve the integration procedure. The performance of the method is evaluated by means of both synthetic and real applications and by comparison with standard integration procedures. The proposed method has proven to lead to improved results, being a convenient way to address the integration problem, when noise is a major source of errors in the signal.     

瞬变电磁采样函数优化法降噪技术

常用的双极性同步采样方法对瞬变电磁系统的观测噪声具有一定的抑制作用,但在实际工作中其噪声抑制能力仍显不足.本文在双极性同步采样方法基础上提出一种优化方法,以提升瞬变电磁系统的噪声抑制能力,即:通过对取样道加高斯窗函数的方式,赋予取样道内各采样点合适的累加权重,提升系统对取样道内高频噪声的抑制能力;使用高斯窗函数与梳状滤波器构造复合窗函数,通过对包含所有叠加周期的完整观测序列加复合窗函数的形式优化各周期的叠加权重,提升系统对实际工频噪声的抑制能力.对优化前后不同采样函数的谱特性及野外实测数据的处理效果进行比较,结果表明本文提出的方法较传统双极性同步采样方法具有更强的噪声抑制能力.     

Site response obtained from array techniques applied to the seismic noise: Two examples in Italy

The vertical component of the seismic noise has been recorded in two different sites near the towns of Mercato S. Severino and Benevento in Southern Italy by a small aperture array, in order to investigate the characteristics of the noise propagation and to study the site response. Three different array techniques have been applied in the two investigated sites: Beam Forming, High Resolution and Spatial Correlation methods. We used two simple array geometry for localising possible noise sources and estimating local shallow structure using ambient noise. The cross shaped array results effective for determining the phase velocity of waves in the case when the noise is from a single localised source; the circular array, on the other hand, is successfully used when the noise sources are distributed. The main results are: the analysis of a coherent component of the noise recorded in the two sites, interpreted as Rayleigh waves, results in reasonable velocity models; the noise recorded in the M.S.S. Plain is a space stationary signal, while the noise at the Benevento site is possibly produced by a stable noise source located close to the array. Due to this evidence, the correlation method does not yield satisfactory results when applied to the Benevento site. The 2–6 Hz spectral peaks of the noise recorded in the M.S.S. Plain can be interpreted as due to a site effect, considering the satisfactory agreement of the noise spectrum at those frequencies with the theoretical transfer function computed on the basis of the velocity model deduced from the Rayleigh waves dispersion analysis.     

自适应滤波方法扣除规则性噪声

本文采用自适应滤波技术,对山东省莒县地震台网记录到的极微震资料中的规则性干扰进行了去噪声处理。结果表明,这种方法不必对噪声进行模式识别,具有简单、快速等优点,在信号与噪声不相关的情况下,可以得到很好的结果。     

IMPULSE RESPONSE MODELS FOR NOISY VIBROSEIS† DATA*

A new method of Vibroseis deconvolution has been recently proposed by the authors. This discussion describes the effects of noise on the application of this method. The initial deconvolution step involves estimating the spectrum of the Vibroseis wavelet by homomorphic filtering. It is shown that noise causes problems with phase estimation. Hence, the Vibroseis wavelet is assumed to be zero phase. Examples demonstrate that zero phase cepstral filtering is a robust wavelet estimation approach for noisy data. The second step of the deconvolution method forms an impulse response model by a spectral extension method. Although this step can improve the resolution of seismic arrivals, it must be applied with caution in view of the deleterious effects of noise.     

基于贝叶斯理论的接收函数与环境噪声联合反演

基于Bayes反演理论(Tarantola,1987,2005),在接收函数非线性复谱比反演方法基础上(刘启元等,1996),本文讨论了接收函数与地震环境噪声Rayleigh波相速度频散的联合反演.本文采用修正后的快速广义反射/透射系数方法(Pei et al., 2008,2009) 计算Rayleigh波相速度频散, 并引入地壳泊松比的全局性搜索.数值检验表明:(1)接收函数与环境噪声的联合反演能够有效地解决反演结果对初始模型依赖的问题,即使对地壳速度结构仅有非常粗略的初始估计(例如,垂向均匀模型),本文方法仍能给出模型参数的可靠估计;(2)由于环境噪声与接收函数在频带上的适配性明显优于地震面波,接收函数与环境噪声的非线性联合反演能更好地约束台站下方近地表的速度结构;对于周期范围为2~40s的环境噪声相速度频散,利用本文方法能够可靠推测台站下方0~80 km深度范围的S波速度结构, 其浅表速度结构的分辨率可达到1 km; (3)本文方法能够可靠地估计地壳泊松比,泊松比的全局性搜索有助于合理解释接收函数和环境噪声的面波频散数据.利用本文方法对川西台阵KWC05台站观测的接收函数与环境噪声的联合反演表明,该台站下方地壳厚度为44 km,上地壳具有明显的高速结构,24~42 km范围的中下地壳具有低速结构.该台站下方地壳的平均泊松比为0.262,壳内低速带的泊松比为0.27.     

Composition of high frequency ambient noise from cross-correlation:A case study using a small aperture array

Long-time cross correlation of ambient noise has been proved as a powerful tool to extract Green's function between two receivers.The study of composition of ambient noise is important for a better understanding of this method.Previous studies confirm that ambient noise in the long period (3 s and longer) mostly consists of surface wave,and 0.25-2.5 s noise consists more of body waves.In this paper,we perform cross correlation processing at much higher frequency (30-70 Hz) using ambient noise recorded by a small aperture array.No surface waves emerge from noise correlation function (NCF),but weak P waves emerge.The absence of surface wave in NCF is not due to high attenuation since surface waves are strong from active source,therefore probably the high ambient noise mostly consists of body wave and lacks surface wave.Origin of such high frequency body waves in ambient noise remains to be studied.     

云南地震数字遥测台网子台地动噪声分析

介绍了云南数字遥测台网的基本情况，阐述了地动噪声有效值和功率谱的计算方法。对选取的资料进行预处理后，计算分析了云南数字遥测台网全部子台的地动噪声水平，按地动噪声水平对子台进行了分类。指出了地动噪声分析对台网建设的作用。     

Hilbert-Huang变换与ELF信号处理

Hilbert-Huang变换(HHT)是近年出现的一种自适应的非平稳、非线性信号处理方法,该方法迄今已在许多非线性、非平稳信号处理的研究领域得到了很好的应用.本文简要介绍了Hilbert-Huang变换的基本实现原理与算法基础,通过仿真信号验证了该方法的有效性,并以实际数据为例,探讨了它在极低频(ELF)信号处理及噪声压制方面的应用.利用Hilbert-Huang变换对ELF信号进行频率域滤波,可以对噪声进行有效压制,从而提取已知频率的电磁信号,极大地提高信号质量.     

Damage reconstruction of 3D frames using genetic algorithms with Levenberg–Marquardt local search

A damage detection method based on harmonic structural vibrations has been applied to reconstruct typical damage patterns of reinforced concrete beams and frames vibrating in three dimensions. It has been shown that comparison of harmonic vibrations of damaged and intact structures makes it possible to formulate optimization criterion which can be solved effectively for large number of unknowns. The application of combined genetic algorithms and Levenberg–Marquardt local search enabled to reconstruct vectors of distinct stiffness distributions containing up to 32 elements with stiffness loss as low as 5% and with accuracy of 1%. To test the algorithm in more realistic environment of practical data acquisition, measuring noise with various levels has been added to the analyzed signal. Such a method of damage detection can be quite effective during post-earthquake assessment of civil engineering structures or if one wants to control the stiffness loss of a reinforced concrete structure after it was accidentally overloaded.     

Feasibility study for an anisotropic full waveform inversion of cross‐well seismic data

Anisotropy is often observed due to the thin layering or aligned micro‐structures, like small fractures. At the scale of cross‐well tomography, the anisotropic effects cannot be neglected. In this paper, we propose a method of full‐wave inversion for transversely isotropic media and we test its robustness against structured noisy data. Optimization inversion techniques based on a least‐square formalism are used. In this framework, analytical expressions of the misfit function gradient, based on the adjoint technique in the time domain, allow one to solve the inverse problem with a high number of parameters and for a completely heterogeneous medium. The wave propagation equation for transversely isotropic media with vertical symmetry axis is solved using the finite difference method on the cylindrical system of coordinates. This system allows one to model the 3D propagation in a 2D medium with a revolution symmetry. In case of approximately horizontal layering, this approximation is sufficient. The full‐wave inversion method is applied to a crosswell synthetic 2‐component (radial and vertical) dataset generated using a 2D model with three different anisotropic regions. Complex noise has been added to these synthetic observed data. This noise is Gaussian and has the same amplitude f?k spectrum as the data. Part of the noise is localized as a coda of arrivals, the other part is not localized. Five parameter fields are estimated, (vertical) P‐wave velocity, (vertical) S‐wave velocity, volumetric mass and the Thomsen anisotropic parameters epsilon and delta. Horizontal exponential correlations have been used. The results show that the full‐wave inversion of cross‐well data is relatively robust for high‐level noise even for second‐order parameters such as Thomsen epsilon and delta anisotropic parameters.     

Making the transition from discrete shot records to continuous seismic records and source wavefields,and its potential impact on survey efficiency and environmental footprint

A marine seismic method based on continuous source and receiver wavefields has been developed. The method requires continuous recording of the seismic data. The source that may consist of multiple source elements can emit signals continuously while moving. The ideal source wavefield to be used with this method should be as white as possible both in a temporal and a spatial sense to avoid deep notches in the spectrum enabling a stable multi-dimensional deconvolution. White noise has such properties. However, equipment that can generate white noise does not exist. In order to generate a continuous source wavefield that is approaching the properties of white noise using existing equipment onboard marine seismic vessels, individual air-guns can be triggered with short randomized time intervals in a near-continuous fashion. The main potential benefits with the method are to reduce the environmental impact of marine seismic surveys and to improve acquisition efficiency. The peak sound pressure levels are significantly reduced by triggering one air-gun at a time compared to conventional marine seismic sources. Sound exposure levels are also reduced in most directions. Since the method is based on continuous recording of seismic data and the air-guns are triggered based on time and not based on position, there are less vessel speed limitations compared to conventional marine seismic data acquisition. Also, because the source wavefield is spread out in time, the wavefields emitted from source elements in different cross-line positions can be designed such that the emitted wavefield is spatially white in this direction. This means that source elements in multiple cross-line positions can be operated simultaneously, potentially improving the cross-line sampling and/or the acquisition efficiency.     

Non-linear three-dimensional inversion of cross-well electrical measurements

Cross-well electrical measurement as known in the oil industry is a method for determining the electrical conductivity distribution between boreholes from the electrostatic field measurements in the boreholes. We discuss the reconstruction of the conductivity distribution of a three-dimensional domain. The measured secondary electric potential field is represented in terms of an integral equation for the vector electric field. This integral equation is taken as the starting point to develop a non-linear inversion method, the so-called contrast source inversion (CSI) method. The CSI method considers the inverse scattering problem as an inverse source problem in which the unknown contrast source (the product of the total electric field and the conductivity contrast) in the object domain is reconstructed by minimizing the object and data error using a conjugate-gradient step, after which the conductivity contrast is updated by minimizing only the error in the object. This method has been tested on a number of numerical examples using the synthetic 'measured' data with and without noise. Numerical tests indicate that the inversion method yields a reasonably good reconstruction result, and is fairly insensitive to added random noise.     

Stochastic analysis of structures and piping systems subjected to stationary multiple support excitations

A stochastic method has been developed for seismic analysis of structures and piping systems subjected to multiple support excitations. In either the time or the frequency domain, mean and extreme values of structural and piping system response can be found, including the effects of cross-correlations of modal response and cross-correlations of multiple support excitations. Stationary white noise and stationary filtered white noise ground excitations are used. A computer program has been developed to carry out the stochastic seismic analysis. Results for a realistic nuclear power plant structure and piping system with and without modal cross-correlations and support excitation cross-correlations are compared. From these results, it is concluded that neglecting cross-correlations can lead to large errors. The stochastic method reported is shown to be more accurate than the response spectrum method and more economical than the time-history method; therefore, it is recommended for seismic analysis of nuclear power plants.     

福建省地震台网预警能力评估

由于常用的均方根值法和噪声功率谱法不能消除不同传感器记录的噪声干扰,为提高噪声水平计算的准确度,本文选用最大概率峰值位移作为背景噪声评估指标。基于可靠的噪声数据,借鉴震级-最大距离监测能力法并考虑预警时效,提出了地震预警最小震级评估方法,系统评估福建三类传感器网及其融合网的地震预警最小震级和预警首报时间。结果表明：测震强震融合网的地震预警最小震级高于单测震网,但明显低于强震网;强震烈度计融合网与单烈度计网的结果相近;三网融合后95%区域的地震预警最小震级约为M_L3.2。由于烈度计网比测震和强震网密集,其预警首报时间最短;三网融合相对于单测震网或单强震网,其震后地震预警首报时间得到了明显缩短,预计95%区域的首报时间为4—6 s。本文研究为福建省的台网布局的优化和重点区域监测能力的提升提供了参考依据。