【作者】 徐善辉；

【导师】 滕吉文； 韩立国；

【作者基本信息】 吉林大学 ， 地球探测与信息技术， 2012， 博士

【摘要】 提高信噪比，是伴随整个地震勘探资料采集处理解释流程的一项重要工作。在高分辨率地震资料的处理工作中，信噪比的提高是根本和前提。而纵观各种已成熟或发展中的地震数据去噪方法，我们可以发现，大部分的地震去噪方法和技术都是建立在通用的数字技术基础上的，这些数字技术往往也可以用在类似的其他领域，如图像图形的去噪增强或压缩算法，语音及机械或其他振动信号的加强或分析等等，并且每种数字技术往往都有多方面的应用。先进的数字信号处理技术有利于在地震反射数据上进行复杂地质目标的成像和分析。经验模态分解(EMD)和Hilbert-Huang变换(HHT)是一个处理非线性非平稳信号的方法。经验模态分解先首先将信号分解为一系列的子信号，这些子信号在经过hilbert变换后会落在相同的时频区域中。HHT为随机信号或伪随机信号提供了一种时频域瞬时属性分析的方法。在本文中，EMD和HHT在地震数据上的应用研究主要包括：1）评价EMD和HHT在时间域及时频域确定有意义的地质信息的能力，2）在地震数据的f-x域内使用EMD分解来压制随机噪声，3）使用HHT的瞬时属性来设计一个好的滤波器，提高地震数据的信噪比。在滤波器设计和实际滤波过程中，HHT所表现出来的经验特性，可以比其他的滤波器更好的提高信噪比。本文使用模拟数据和实际地震数据证明了HHT和EMD可以成功应用在地震反射数据的处理上。在其中的一个实验中，强烈的工业交流电噪声可以通过HHT的陷频滤波进行消除。在另一实验中，使用HHT方法进行面波的压制，与其它方法相比，显示其具有较好的去噪效果。本篇文章的主要研究工作是如何更好的将EMD和HHT技术应用在地震反射数据上，尝试将EMD，HHT在其他领域中的新特点使用在地震勘探上，以提高反射地震数据的质量。这些研究不是要取代传统的地震数据处理方法，而是要将EMD和HHT补充到其中去。EMD是一种非常有效的时间域滤波器，HHT又是一种可以同时在时间和频率方面分析信号的优秀方法。同时，HHT的最大优点是它可以不通过任何先验信息就可以获取信号的瞬时振幅，瞬时频率和瞬时相位信息。先验信息通常是指采样率，噪声尖峰及步函数等时间序列的局部分布。本文首先介绍了EMD分解的有关基本概念及实施过程，以一个由多频率简谐波合成的调频信号为例来说明EMD分解的具体做法，相关概念，分解后的结果及原始信号的重构方法等。随后讨论了EMD分解应该注意的问题及EMD分解的不当会产生的问题，EMD分解的一个缺陷是容易产生频率混叠及出现原信号中不包含的虚假IMF分量，本文提出使用可变阈值的方法来改善这一缺陷，同时，阈值方法的改进也大幅加快了EMD分解的效率。为了讨论f-x域内的地震数据去噪方法，叙述了f-x反褶积的基本原理及发展过程，并讨论了其局限性，由此引出了在f-x域内使用EMD滤波方法来消除地震噪声的观点。使用模拟数据加入噪声前后的f-x域谱等频信息的对比来说明EMD滤波压制噪声的可行性。提出了f-x域EMD与小波阈值法的联合去噪的方法。使用简单模型，复杂模型的模拟数据来验证这种方法的有效性并与其他去除随机噪声的方法比较说明其优缺点。随后使用叠前及叠后的实际地震数据进行去噪实验，并对去噪结果进行详细分析。本文所提出的f-x域EMD与小波阈值法联合噪声衰减的方法是Battista B.M的f-x域EMD去噪方法进行了改进，通过分析EMD分解的过程指出直接删除IMF1分量的去噪方法是不严格的，而本文所提出的f-x域EMD与小波阈值法联合噪声衰减的方法在有效衰减噪声的同时可以更大程度的保留有效信号成分，整体效果优于f-x反褶积等其他去噪方法。本文系统的将HHT滤波技术引入到地震数据规则噪声的衰减过程中来。具体包括：基于HHT的工频噪声衰减，基于HHT的面波衰减方法研究和基于HHT的海上地震数据涌浪噪声衰减方法研究。首先分析了这几种规则噪声的特点，它们在EMD分解及Hilbert-Huang变换的过程中是易于分离的，从中得出结论，适合使用HHT滤波的方法来衰减这几种地震数据中的规则噪声。对模拟和实际地震数据使用所设计的方法进行去噪测试，结果表明基于HHT的方法与基于FFT的方法相比可以更好的压制噪声并保留有效信号。本文简要介绍了去除地震数据中工业交流电及面波干扰的常规方法，并讨论了这些方法的优缺点。随后，简要介绍了Hilbert-Huang变换的基本过程，并随即给出了基于HHT的时频域滤波的概念及实施过程。EMD和HHT对工频噪声具有突出的分离和时频分析的能力，因此可以设计一种使用HHT进行陷频滤波以消除地震数据中的工频噪声干扰的方法。对含工频噪声的地震数据进行EMD分解及HHT的计算，结果表明EMD分解可以将工频噪声较为完整的分离出来，进而在HHT谱上有很好的对应于工频噪声的时频显示。对单道含工频噪声的地震信号进行HHT陷频滤波去噪实验，结果表明该方法可以有效的压制噪声并保留有效信号。与其他基于FFT的滤波方法比较，该方法的去噪效果是最好的。随后对实际数据进行去噪实验，并与普通陷频滤波方法进行对比。本文进行了HHT时频滤波对面波干扰的压制研究。对合成含面波数据及实际数据都进行了HHT分析，所得时频谱中可以清楚的得到面波所对应的时频信息，对实际数据使用HHT滤波方法及FK滤波两种方法来压制噪声，对所得结果进行对比分析，本文方法优于传统的FK面波压制方法。而对于海上地震数据通常包含的低频强振幅的涌浪噪声，进行了HHT滤波方法压制研究。通过对实际数据的分析指出，直接使用EMD分解的方法来剔除涌浪噪声是不合适的，针对实际数据的涌浪噪声的特点，本文设计了HHT低切滤波的方法，有效的滤波除了海上数据的涌浪噪声，与FFT低切滤波方法的去噪效果进行对比分析，结果表明本文方法在有效压制噪声的同时更大程度的保留了有效反射信号。本文对基于BEMD的二维去噪技术及地震属性提取技术进行了研究，首先简要介绍了BEMD分解的基本过程，包括二维数据局部极值点求取方法，构造包络曲面的插值计算方法等。随后介绍了目前BEMD分解技术还存在的问题及改进方向。使用BEMD技术对二维双曲面数据进行滤波，与中值滤波，均值滤波，维纳滤波及小波阈值滤波方法比较，BEMD滤波对该数据的噪声消除效果更好。将BEMD技术引入到地震属性提取及相干数据体上断层信息的提取。BEMD对二维信号高低频分离的特性可以用于识别时间切片上的隐伏圈闭等构造。BEMD同时具有很好的边缘检测和纹理识别的能力，对相干数据的切片进行BEMD处理可以压制非构造的噪声信号，将其用于提高在地震相干数据切片上的断层识别能力，使断层信号更清晰连续，效果很好。对数谱在常规地震勘探数据处理中的使用方法主要是反褶积及子波提取，本文进行了基于对数谱的薄层估计研究。在数字信号处理的其他领域，如音频信号的处理，对数谱（倒谱）的定义方法将信号频谱的自然对数作为信号再求其频谱。本文分析了这种对数谱的特点，首先研究如何使用对数谱来快速测试地震仪正弦波信号的畸变。对数谱应用在地震仪信号畸变程度的检测当中时，可以更快速直观的判断正弦波信号的畸变程度。使用对数谱可以进行薄层厚度预测，所得的关于薄层厚度的分辨率要高于直接在时间域及频率域的厚度识别能力，本文使用楔状模型试验来检验对数谱在薄层识别上的效果，信号主频在30Hz时，薄层的识别能力可以达到10ms以内。本文提出了一种使用对数谱进行低频信号检测的方法并与其他谱分解方法进行比较，结果表明对数谱方法在对低频信号检测过程中，所取得的时间和频率分辨能力要优于其他的谱分解的方法。更多还原

【Abstract】 The improving of signal-to-noise ratio is always an important task accompaniedby the entire seismic data processing and interpretation. In the high-resolutionseismic data processing, the improving of signal-to-noise ratio is the fundamentaland premise. A general review of various maturity or development of seismic datadenoising method, we can find that, most of the seismic denoising methods andtechniques are built on a basis of common digital technology, these digitaltechnologies often used in similar in other areas, such as enhancement orcompression algorithm of image and graphics, the strengthen or analysis of voiceand mechanical vibration signal, etc., and each digital technology often have a widerange of applications.Advanced digital signal processing technology is conducive tothe imaging and analysis of seismic reflection data with complex geological targets.At present a method to handle non-linear non-stationary signals is the empiricalmode decomposition (EMD) and Hilbert-Huang transform (HHT). First of all, thesignal is decomposed into a series of sub-signal by the empirical modedecomposition, these sub-signals would fall on the same frequency region afterhilbert transform.HHT provides an instantaneous attributes of time and frequencydomain analysis method for random signal or pseudo-random signal. In this paper,the main application of EMD and HHT on the seismic data is:1) evaluate the abilityto identify meaningful geological information of the EMD and HHT in a timelymanner in the time domain to frequency domain,2) Use of EMD decomposition tosuppress random noise in the the f-x doman of the seismic data,3) design a betterfilter to improve the seismic data signal-to-noise ratio by the instantaneous attributesof HHT. In filter design and the actual filtering process, the experience characteristicof the HHT demonstrating, can improve the signal-to-noise ratio better than anyother filte.In this paper we use the simulated data and real seismic data to prove thatthe HHT and EMD can be successfully used in the processing of seismic reflectiondata. In an experiment, a strong industrial AC noise can be eliminated by HHT notchfilter. In another experiment, using the HHT method, surface wave is eliminated,compared with other methods, its better denoising is shown. The main research work of this paper is how to use the EMD and HHT technology applications better inseismic reflection data, try to use the new features of EMD and the HHT in otherareas in the seismic exploration to improve the quality of seismic reflection data.These studies not intended to replace the traditional seismic data processing methods,but rather to supplement the EMD and HHT technology to the processing methods.EMD is a very efficient time-domain filter, the HHT is a good way to analysissignals in terms of time and frequency. Meanwhile, the biggest advantage of HHT isthat it can provide you the instantaneous amplitude the instantaneous frequency andinstantaneous phase information of the signal without any priori information. Prioriinformation usually refers to the sampling rate, the local distribution of time seriessuch as the noise peak and step functions.In this paper, we first introduced the basic concepts and implementation processof the EMD. A signal which is mixed by some harmonic with different frequency istaken for example to illustrate the EMD practices, related concepts, the result ofdecomposition and the original signal reconstruction method. Then we discussed theproblems in EMD decomposition and problems should be noted when the improperEMD decomposition will produce. In order to discuss the f-x domain seismic datadenoising method described f-x deconvolution of the basic principles anddevelopment process, and discusses its limitations, which leads to the view that weshould use the EMD filtering method in the f-x domain to eliminate seismic noiseUsing simulated data with and without noise adding, by comparing thecross-frequency information in the f-x domain to illustrate the feasibility of EMDfilter to suppress noise.F-x domain A EMD and wavelet thresholding joint denoisingmethod was put forward. Using simulated data of simple models and complexmodels we verified the effectiveness of this method and by comparison with othermethods to remove random noise we explained the advantages and disadvantages.Subsequent a denoising test is taken by the actual prestack and poststack seismicdata with random noise, then detailed analysised the denoising results.In this paper, the joint noise attenuation method of the f-x domain EMD andwavelet thresholding is improved from the Battista, BM’s EMD denoising method inf-x domain,by the analysis of EMD process we can get a conclusion that directlydeleting IMF1component denoising method is not strictly, this proposed the f-xdomain EMD and wavelet thresholding joint noise attenuation greater degree ofeffective noise attenuation at the same time to retain effective signal components, the overall effect is better than f-x deconvolution and other denoising methods.The HHT filter technology was first introduced into the process in the regularnoise attenuation of seismic data in this article. Including: HHT-based powerfrequency noise attenuation, based on HHT surface wave attenuation methods andHHT-based marine seismic data swell noise attenuation. First analysised thecharacteristics of these types of noise, they are easily separated in the process ofEMD and Hilbert-Huang Transform, and draw conclusions from which that it issuitable to attenuation these kinds of regular seismic data noise for HHT filtermethod. Results show that the HHT-based approach compared to the FFT-basedapproach can better suppress noise and to retain effective signal noise tests onsimulated and real seismic data using the design method.This article briefly introduced surface wave interference with conventionaldenoising methods, and then discusses the advantages and disadvantages of thesemethods.Subsequently, a brief introduction to the basic process of the Hilbert-Huangtransform, and then gived the concept and implementation process of thetime-frequency domain filtering which is based on HHT. Based on the outstandingability of separation and time-frequency analysis of EMD and HHT,we design anotch filter to eliminate the power frequency noise in seismic data based on HHTtechnology. The results of EMD and HHT calculation of seismic data includingpower frequency signal show that in the process of EMD decomposition powerfrequency signal has a more complete separation and a well corresponds to thefrequency signal frequency display on the HHT spectrum. The experimental resultsof single channel containing the power frequency noise of seismic signals to HHTnotch filter denoising show that this method can effectively suppress noise and retaineffective signal. Compared with the FFT-based filtering method, the denoising effectof this method is the best. Then a experiment by the actual data noise was done,frequency filtering methods were compared with the ordinary trap. In this article aresearch of suppression of wave interference on HHT time-frequency filteringopposite was carried out. Synthesis and actual data of surface wave were andanalysis HHT spectrum obtained when the surface waves corresponding to thetime-frequency information, the actual data using the the HHT filter method and FKfiltering are two ways to suppress noise, comparative analysis of the results of thismethod due to the FK surface wave suppression methods. The section3.5is aboutthe study of marine seismic data the HHT filtering method to suppress swell noise. Through the analysis of actual data that directly using the EMD method to weed outthe swell noise is not appropriate, the characteristics of surge noise of the actual data,this paper designs a of HHT low cut filter method, the effective filtering in additionto the at-sea data the noise of the surge, were analyzed with FFT low-cut filterdenoising results show that this method effectively suppress noise while a greaterdegree of retained effectively reflected signal.Two-dimensional de-noising technology and seismic attributes based on BEMDextraction technology is studied. Firstly briefly describes the basic process of BEMDdecomposition, including how to get the local extreme point of the two-dimensionaldata and how to calculate the structural envelope surface through the interpolationcalculation. Then introduced some existing problems in the currently BEMDdecomposition technique and the improving direction. BEMD is used in the filteringof two-dimensional surface data, comparing with other filtting method such an themedian filtering, mean filtering, wiener filtering and wavelet threshold filtering,noise cancellation of the BEMD filtering of the data is the better. BEMD technologyis introduced into the extraction of seismic attribute and the fault information in thecoherent data body. BEMD’s separation characteristics of the low and highfrequency in two-dimensional signal can be used to identify seismic structure such asinsidious trap on the time slices. BEMD also has the ability of edge detection andtexture recognition, slices of coherent data BEMD treatment to suppress thenon-structural noise signal to be used to improve the ability to identify faults inseismic coherence data slice, with good results. Fault signal more clearly continuous,fault identification capability in the coherent data.The application of logarithm spectrum in the conventional seismic dataprocessing is deconvolution and wavelet extraction, in this paper, logarithmspectrum is used to estimate the thickness of thin layer In other areas of digital signalprocessing, such as the audio signal processing, the logarithm spectrum (cepstrum) isdefined signal spectrum of the natural logarithm of the logarithm spectrum and thenseek its spectrum. This paper analyzes the spectrum characteristics of the logarithmspectrum, first of all how to use the logarithmic spectral to test the distortion of thesine wave signal generation by a seismograph. Logarithm spectrum used in thedetection of the seismograph signal distortion of which can judgment sine wavedistortion of the signal more quickly and intuitively. Logarithmic spectrum can beused in a thin layer thickness predicting when the resolution is higher than that obtained on the thin layer thickness directly in the time domain and frequencydomain, the thickness of the ability to identify, we use the wedge model test to verifythe identification effect of the thin layer using the logarithmic spectrum, when thedominant frequency of the signal is30Hz, the recognition ability of the thin layer canbe less than10ms. In this paper, a low-frequency signal detection method usinglogarithmic spectra is put forward, comparison with other spectral decompositionmethods. Finally the results show that the capability of logarithmic spectrum in thelow-frequency signal detection process is better comparing with other spectraldecomposition method, in the time and frequency resolution.更多还原