【作者】 姚建红；

【导师】 高丙坤；

【作者基本信息】 大庆石油学院 ， 油气信息与控制工程， 2010， 博士

【摘要】 提高地震信号的信噪比是提高地震信号分辨率的先决条件,要提高地震资料的信噪比,需要去除地震资料中的噪声。地震勘探信号中的噪声主要有两种:相干噪声和随机噪声。相干噪声在时间上具有规律性,可以依据这些规律加以去除,一般相干噪声主要包括面波、多次波、折射鸣震等。而随机噪声没有统一的规律,它包括环境噪声、测量误差、地面微震等。本文主要研究了随机噪声的去除。在很多地震信号去噪的方法中都用到傅立叶变换和短时傅立叶变换,但傅立叶变换只对于确知信号和平稳随机过程有显著的意义,而实际信号往往是时变信号、非平稳过程,了解它们的局部特性常常是很重要的,短时傅立叶变换又不能同时兼顾时间分辨率和频率分辨率。近年来,小波变换在地震信号处理中的应用一直是人们研究的一大热点,己取得了令人瞩目的成果。因为它在时域和频域都具有很好的局部化性质,可以将信号所携带的信息分解到任意细节加以分析,并且信号和噪声在小波变换的细节信息具有截然不同的特性,因此可将其用于地震信号去噪,以更为准确地区分有效信号和噪声,在最大限度去除噪声的同时,尽可能保留有效信号。本论文主要利用离散二进小波变换对地震信号进行分析和去噪处理。论文首先介绍了地震波场的特点,对地震信号中主要噪声类型进行介绍,并针对其特点介绍了目前使用的去除相干噪声和随机噪声的各种方法,并简要分析了各方法的优缺点。论文对小波变换基本原理做了较为详细的介绍,讨论了小波变换的基本原理及常用小波函数。针对信号和噪声在小波变换下模极大值在各尺度上不同的传播特性,利用小波变换模极大值进行信号去噪和重构。基于小波变换模极大值的地震信号去噪方法能够较好地去除地震信号中随机噪声,且在不同尺度上能够更精细地区分信号和噪声,使得在去除高频噪声的同时,可以尽可能多地保留有效的高频信息。由于白噪声具有负的奇异性,且其幅度和稠密度随尺度增加而减少,因此如果某个信号的小波变换局部模极大值的幅度及稠密度随尺度减小而快速增大,表明该处的奇异性主要由噪声控制,在去噪时应予去除。因此文中采用了一种有噪声控制的模极大值的算法来进行消噪,取得了较好的处理结果。论文介绍了常规小波域阈值去噪处理方法,并分析了小波阈值去噪的性能。小波阈值去噪虽然能够去除信号中随机噪声,提高信噪比,但在实际应用中单独使用常规的小波域阈值去噪,效果有时还不能令人满意。因此文中采用了一种结合软硬阈值的加权阈值去噪方法,经实际地震剖面处理验证,该方法比直接应用常规的小波域阈值去噪有更好的处理效果,对信号的逼近程度高,并能抑制Gibbs振荡现象。论文针对小波变换模极大值的信号去噪方法存在算法稳定性较差及计算量很大的问题,讨论了另一种基于小波变换在不同尺度传播特性的去噪方法——基于小波变换尺度间相关性去噪方法。通过把尺度间的小波系数进行相关处理,则有效信号的小波系数被相关加强,而噪声的小波系数被相关减弱,以此来实现信噪分离。论文分析了同一尺度下的迭代过程、不同类型和不同信噪比下含噪信号的去噪效果,并研究了一种改进的尺度相关去噪算法,经对实际地震信号处理结果表明了其有效性。在论文的最后,总结了本文所做的工作和成果,指出了研究的不足之处和将来的研究方向。更多还原

【Abstract】 The prerequisite condition of improving the resolving power is to improve the SNR of signal, which need to remove the noise from seismic data. There are two kind of noise in seismic data: coherent noises and random noises. Coherent noises are disciplinary and can be removed according to the rules. Usually, coherent noises comprise of surface wave, multiple wave, refraction and ringing. Random noises have uniform rules and they comprise of circumstance noise, measurement error and ground microseism. The paper discusses the random noise removing.Fourier transformation and short Fourier transformation are used in many seismic data noise removing. But Fourier transformation is applicable to deterministic signal and stationary and related stochastic processes. The practical signal usually is time-varied and non-stationary and it is important to comprehend its local characteristic. Short Fourier transform can’t consider time resolution and frequency resolution. Recently, wavelet transform has become hotspot in seismic signal process and many fruits have been acquired. Wavelet transform has good local characteristic in time and frequency domain and can decomposes arbitrary detail of signal.The detail characteristic of signal and noise are different during wavelet transform, which can be used to remove noise of seismic data in order to distinguish effective signal and noise. The effective signal can be reserved when the noises are removed as far as possible. The dyadic discrete wavelet transform is used to analyze seismic data and remove noise.The characteristic of seismic wave scene is introduced first, then the main noise styles in seismic data are discussed. The noise removing method of coherent noise and random noise are also introduced with their characteristics. Then, the paper discusses the discipline of wavelet transform and several wavelet functions. As the different propagation characteristic of signal and noise on different scales of modules maxima, the signal can be de-noised and reconstructed. The method can remove random noise in seismic data preferably, and the signal and noise can be distinguished on different scales, which can reserve the effective high frequency information as far as when removing high frequency noise. The white noise has native singular, its amplitude denseness decrease with scales, so if the amplitude and denseness of local modules maxima for a signal decrease quickly with scales, it indicates that the singular of this point is controlled by noise and should be removed. Therefore, a modules maxima algorithm controlled by noise is adopted in the paper to remove noise and acquires prefer results.The general de-noising method based on wavelet threshold is also introduced and its performance is analyzed. Although the threshold method can remove random noise, the effect couldn’t turn up trumps. So the weighted threshold combined soft and hard-threshold is adopted in the paper, the process results demonstrate that it can approach to signal preferably and restrain Gibbs oscillation.The noise removing based on modules maxima of wavelet transform has the problem such as bad stability and large calculation, so inter-scale relativity de-noising method is discussed. The wavelet coefficients are enhanced relatively by correlating the inter-scale coefficients to distinguish the signal and noise. The paper analyzes the noise removing effect such as iterative course on same scale, different style and SNR, and an improved inter-scale relativity de-noising algorithm is discussed. The results demonstrate its effectiveness.At last, the work and harvest is summarized, the deficiency and future direction are also indicated.更多还原