摘要
K均值序贯泛化(SGK)去噪算法在字典更新阶段会引入噪声干扰。为了控制噪声干扰对字典原子的影响,构建一种干扰控制K均值序贯泛化(C-SGK)地震信号去噪算法。该算法在字典更新阶段通过判断信噪比值与设定阈值间的大小来决定是否更新原子。若信噪比值大于设定阈值,则顺序更新原子,反之则不更新原子。对人工合成和实际地震信号的去噪结果表明,本算法能够很好地控制噪声干扰,且与传统SGK算法比较发现,本文算法对地震信号的去噪效果更优。
A certain amount of noise would be introduced by a dictionary update step using the K-means sequential generalized(SGK)denoising algorithm.To reduce the effect of noise interference on dictionary atoms,a seismic signal denoising algorithm is proposed based on controlled interference SGK(C-SGK)dictionary learning under a compressive sensing framework.The algorithm compares the signal-to-noise ratio and the threshold set in the dictionary update step,which determines whether to update the atom:the atoms should be sequentially updated only if the signal-to-noise ratio is greater than the threshold.The experimental results of synthesized and real seismic signal denoising in this study indicate that the proposed algorithm can effectively control noise interference.Compared with traditional SGK denoising, the proposed algorithm demonstrates a better denoising effect on seismic signals.
引文
[1]Zhang H, Chi Y, Zhou Y T,et al. Threedimensional seismic signal denoising based on four-dimensional block matching cooperative filteringcombined with principle component analysis[J].Laser &Optoelectronics Progress,2018,55(4):041007.张欢,池越,周亚同,等.结合主成分分析的四维块匹配协同滤波三维地震信号去噪[J].激光与光电子学进展,2018,55(4):041007.
[2]Zhang M,Lv X Q, Wu L,etal.Multiplicativedenoising method based on deep residual learning[J].Laser &Optoelectronics Progress,2018,55(3):031004.张明,吕晓琪,吴凉,等.基于深度残差学习的乘性噪声去噪方法[J].激光与光电子学进展,2018,55(3):031004.
[3]Ni X,LüQ W,Meng F,et al.Image denoisingmethod based on curvelet transform and totalvariation[J].Acta Optica Sinica,2009,29(9):2390-2394.倪雪,李庆武,孟凡,等.基于Curvelet变换和全变差的图像去噪方法[J].光学学报,2009,29(9):2390-2394.
[4]Yu S,Cai X,Su Y.Seismic signal enhancement bypolynomial fitting[J].Applied Geophysics,1989,1(1):57-65.
[5]Zhu W H,Kelamis P G,Liu Q L.Linear noiseattenuation using local radial trace median filtering[J].The Leading Edge,2004,23(8):728-737.
[6]Li Y J,Yang B J,Li Y,et al.Combining SVD withwavelet transform in synthetic seismic signaldenoising[C]∥Proceedings of 2007 InternationalConference on Wavelet Analysis and PatternRecognition,2007,4:1831-1836.
[7]Tang G,Ma J W,Yang H Z.Seismic data denoisingbased on learning-type overcomplete dictionaries[J].Applied Geophysics,2012,9(1):27-32.
[8]Zhang G Z,Chang D K,Wang Y H,et al.3Dseismic random noise suppression with sparse andredundant representation[J]. Oil GeophysicalProspecting,2015,50(4):600-606.张广智,常德宽,王一惠,等.基于稀疏冗余表示的三维地震数据随机噪声压制[J].石油地球物理勘探,2015,50(4):600-606.
[9]Tang J Y,Chen W T,Chen S Y,et al.Applicationof wavelet-based denoising using a new adaptivethresholding function to vibration signals[J].Journalof Vibration and Shock,2009,28(7):118-121.唐进元,陈维涛,陈思雨,等.一种新的小波阈值函数及其在振动信号去噪分析中的应用[J].振动与冲击,2009,28(7):118-121.
[10]Wang H Q,Shang C Y,Gao R P,et al.Animprovement of wavelet shrinkage denoising viawavelet coefficient transformation[J].Journal ofVibration and Shock,2011,30(10):165-168.王宏强,尚春阳,高瑞鹏,等.基于小波系数变换的小波阈值去噪算法改进[J].振动与冲击,2011,30(10):165-168.
[11]Chen G Y,Kégl B.Image denoising with complexridgelets[J].Pattern Recognition,2007,40(2):578-585.
[12]Ji J,Xu S X,Li X.An adaptive thresholding imagedenoising method based on morphological componentanalysis and contourlet transform[J]. PatternRecognition and Artificial Intelligence,2014,27(6):561-568.纪建,许双星,李晓.基于形态成分分析和Contourlet变换的自适应阈值图像去噪方法[J].模式识别与人工智能,2014,27(6):561-568.
[13]Bhadauria H S,Dewal M L.Medical image denoisingusing adaptive fusion of curvelet transform and totalvariation[J].Computers &Electrical Engineering,2013,39(5):1451-1460.
[14]Yang N,He Z H,Huang D J.Signal and noiseseparation method for pre-stack seismic data in waveatomic domain based on coefficient correlationthreshold[J].Oil Geophysical Prospecting,2011,46(1):53-57.杨宁,贺振华,黄德济.基于系数相关性阈值的波原子域叠前地震资料信噪分离方法[J].石油地球物理勘探,2011,46(1):53-57.
[15]Elad M,Starck J L,Querre P,et al.Simultaneouscartoon and texture image inpainting usingmorphological component analysis(MCA)[J].Applied and Computational Harmonic Analysis,2005,19(3):340-358.
[16]Mairal J,Bach F,Ponce J,et al.Non-local sparsemodels for image restoration[C]∥Proceedings of2009IEEE 12th International Conference on ComputerVision,2009:2272-2279.
[17]Mairal J,Sapiro G,Elad M.Learning multiscalesparse representations for imageand video restoration[J].Multiscale Modeling &Simulation,2008,7(1):214-241.
[18]Quan Y L,Zhu T Y,Harris J M,et al.Imageintegration with learned dictionaries and application toseismic monitoring[C]∥Proceedings of SEGTechnical Program Expanded Abstracts,2011,30(1):4217-4222.
[19]Cai J F,Ji H,Shen Z W,et al.Data-driven tightframe construction and image denoising[J].Appliedand Computational Harmonic Analysis,2014,37(1):89-105.
[20]Protter M,Elad M.Image sequence denoising viasparse and redundant representations[J].IEEETransactions on Image Processing,2009,18(1):27-35.
[21]Bryt O,Elad M.Compression of facial images usingthe K-SVD algorithm[J]. Journal of VisualCommunication and Image Representation,2008,19(4):270-282.
[22]Zibulevsky M, Pearlmutter B A. Blind sourceseparation by sparse decomposition in a signaldictionary[J].Neural Computation,2001,13(4):863-882.
[23]Sahoo S K,Makur A.Dictionary training for sparserepresentation as generalization of K-meansclustering[J].IEEE Signal Processing Letters,2013,20(6):587-590.
[24]Sahoo S K,Makur A.Enhancing image denoising bycontrolling noise incursion in learned dictionaries[J].IEEE Signal Processing Letters,2015,22(8):1123-1126.
[25]Chen Y K. Fast dictionary learning for noiseattenuation of multidimensional seismic data[J].Geophysical Journal International,2017,209(1):21-31.
[26]Hennenfent G, Herrmann F J.Simply denoise:wavefield reconstruction via jittered undersampling[J].Geophysics,2008,73(3):V19-V28.
[27]Herrmann F J, Hennenfent G. Non-parametricseismic data recovery with curvelet frames[J].Geophysical Journal International,2008,173(1):233-248.