【作者】 李晓璐；

【导师】 周亚同；

【作者基本信息】 河北工业大学 ， 电子与通信工程（专业学位）， 2020， 硕士

【摘要】 地震勘探是石油勘探的重要手段,由地表激发产生的地震数据可以反演地下地质结构及石油蕴藏情况,而采集的地震数据含有大量噪声,不利于后续数据分析。非局部均值算法(Non-Local Means,NLM)自提出以来,被广泛用于彩色图像降噪并获得让人满意的效果,而在地震数据降噪领域的应用却仍需深入研究。本文充分分析含噪地震数据的特点,以NLM为框架开展对地震数据的降噪研究。主要研究内容如下:(1)融合边缘检测的非局部均值地震数据降噪针对NLM在地震数据降噪中出现的同相轴被平滑的问题,提出一种融合边缘检测的非局部均值地震数据降噪算法(Sobel8-NLM)。首先用八方向Sobel算子检测含噪地震数据的同相轴,再用检测到的同相轴改进NLM的权值,使相差较大的邻域权值减小,在实现降噪的同时,较好地恢复地震数据中的同相轴。实验以合成、海上和陆上地震数据为样本,与NLM、结合经典Sobel算子的非局部均值算法(Sobel2-NLM)、结合Canny算子的非局部均值降噪算法做对比,验证算法在地震数据降噪中的优势。(2)基于自适应快速改进非局部均值的地震数据降噪针对Sobel8-NLM存在滤波参数无法随噪声大小发生适应性变化、计算量较大这两个问题,提出一种自适应快速改进非局部均值的地震数据降噪算法。对含噪地震数据采取两次Sobel8-NLM降噪,并改进滤除参数,同时用互相关函数替代卷积运算,在提高地震数据降噪效果的同时减少运算时间。实验以合成、海上和陆上地震数据为样本,对比NLM,Sobel8-NLM和基于最近邻选择策略的非局部均值算法的降噪效果和降噪时间,验证了算法对地震数据的降噪性能。最后通过对野外采集到的地震数据做降噪实验,证明算法具有鲁棒性。(3)基于全变分正则化非局部均值的地震数据降噪全变分正则化约束可以有效保持地震数据中同相轴,为此提出一种基于全变分正则化非局部均值的地震数据降噪算法。利用NLM对地震数据的降噪结果,更新权值以去除抖动效应,对更新权值后的NLM降噪结果进行全变分正则化约束。实验以合成、海上和陆上地震数据为样本,并与NLM,基于最近邻选择策略的非局部均值算法的降噪效果进行比较,并在最后对比算法效率,验证了算法在地震降噪中的优势。更多还原

【Abstract】 Seismic exploration is an important means of petroleum exploration.Seismic data generated by surface excitation can be used to invert the underground geological structure and oil reserves.However,the collected seismic data contain a large amount of noise,which is not conducive to subsequent data analysis.Non-local means(NLM)algorithm has been widely used in color image denoising and has achieved satisfactory results since it was proposed.However,its application in seismic data denoising still needs further study.In this paper,the characteristics of noisy seismic data are fully analyzed,and the research on noise reduction of seismic data is carried out based on NLM framework.The main research contents are as follows:(1)Non-Local Means Seismic Data Denoising by Combination of Edge DetectionAiming at the problem that the events are smoothed when the non-local means algorithm is used for seismic data denoising,a non-local mean seismic data denoising algorithm(Sobel8-NLM)based on edge detection is proposed.Firstly,the eight-directions Sobel operator is used to detect the events of noisy seismic data,and then the detected events information is used to improve the weight of NLM,so that the neighborhood weights with large difference are reduced,and the events of seismic data is well recovered while denoising is realized.The experiment takes synthetic,marine and land seismic data as samples,and compares them with NLM,non-local means denoising algorithm(Sobel2-NLM)combined with classical Sobel operator and non-local means noise reduction algorithm combined with Canny operator to verify the advantages of the algorithm in seismic data denoising.(2)Seismic Data Denoising Based on Adaptive Fast Improvement of Non-local MeansAiming at the two problems of Sobel8-NLM,that the filtering parameter cannot change adaptively with the noise level and the calculation amount is large,an adaptive fast improved non-local means seismic data denoising algorithm is proposed.Sobel8-NLM noise reduction is applied to noisy seismic data twice,filtering parameters are improved,and cross-correlation function is used instead of convolution operation to improve the noise reduction effect of seismic data and reduce operation time.The experiment takes synthetic,marine and land seismic data as samples,compares the noise reduction effect and time of NLM,Sobel8-NLM and non-local means algorithm based on nearest neighbor selection strategy,and verifies the denoising performance of the algorithm on seismic data.Finally,through the noise reduction experiment of the seismic data collected in the field,the robustness of the algorithm is verified.(3)Total Variational Regularization Based on Non-local Means Seismic Data Denoising Total variation regularization constraint can effectively maintain the events in seismic data.Therefore,a total variation regularization based on non-local means seismic data denoising algorithm is proposed.Using the NLM denoising results of seismic data,the weights are updated to remove the jitter effect,and the NLM noise reduction results after updating the weights are subjected to total variational regularization constraints.The experiment takes synthetic,marine and land seismic data as samples,and compares the noise reduction effect of NLM,the non-local means algorithm based on the nearest neighbor selection strategy,and the efficiency of the algorithm is compared at the end,which verifies the advantages of the algorithm in seismic denoising.更多还原