Research On Sparse Denoising Of KSVD Based On CEEMD

When seismic data are collected,they are susceptible to different environmental factors.There are many random noises in seismic data,which lead to a decrease of the resolution and signal-to-noise ratio of seismic profiles.The quality of the data directly affects the accuracy of geological interpretation.The better the quality,the more accurately the true situation of underground geology can be revealed.In order to improve the quality of data processing,the removal of random noise is an indispensable part.In this paper,a lot of detailed research work has been done on the removal of random noise in seismic data,in order to find a more efficient and practical denoising method.Firstly,the article studies the basic theory of empirical mode decomposition(EMD),its properties and several shortcomings of the algorithm itself.Aiming at the shortage of EMD,two algorithms have been developed.One is Ensemble Empirical Mode Decomposition(EEMD),which solves the problem of EMD to a certain extent.However,the decomposed IMF can not keep the definition of IMF accurately and its completeness is not good.The second is Complete Ensemble Empirical Mode Decomposition(CEEMD).The results of CEEMD are better than those of EMD and EEMD,but there are still some problems.The value of the second mode always tends to zero at any time in the whole time period.Meanwhile,with the addition of Gauss white noise,there will be more or less noise interference in the decomposed mode.In order to solve the above problems,CEEMD was improved.Comparing the decomposition results of theoretical synthetic signals by EMD,EEMD and CEEMD before and after improvement,the improved CEEMD decomposition is the best,the error is the smallest,and the calculation efficiency is the highest,so it has a good application prospect.Then,the theory of sparse representation is briefly described in this paper,including the mathematical model of sparse representation,the development of dictionary construction,and the commonly used sparse decomposition algorithm.The learning dictionary(KSVD dictionary)and the sparse denoising method based on KSVD dictionary are emphatically introduced.when the algorithm of KSVD dictionary is used to obtain the dictionary,there is a problem that the training time takes too long,which leads to the double sparse KSVD dictionary algorithm.The method can not only accurately describe the characteristics of signals,but also calculate faster than the KSVD dictionary algorithm.By comparing the denoising results of synthetic seismogram data,it is proved that the sparse denoising method based on double-sparse KSVD dictionary has a relatively good effect on signal random noise removal.Finally,this paper carefully studies and analyzes the advantages and disadvantages of CEEMD frequency division denoising and KSVD sparse decomposition denoising.Based on the advantages of both,the improved CEEMD and the double-sparse KSVD dictionary sparse denoising method are organically combined.and the proposed sparse denoising of KSVD based on CEEMD is really good in denoising the model data and actual data.The signal-to-noise ratio is improved,the quality of seismic data is greatly improved,and its application to the identification of low-order faults has also got a good practical result.