The Research Of Seismic Data Denoising And Separation For Multi-source Blended Seismic Data Via Sparse Representation

Seismic data acquisition requires a trade-off between quality and cost.The traditional acquisition method adopts the method of firing multi-point receiver one shot at a time.Multi-source hybrid acquisition is an efficient seismic acquisition method,which can avoid the shortcomings of the traditional acquisition method.Multi-source acquisition can obtain seismic records containing multiple wave fields by simultaneously stimulating multiple sources after the source is coded in a certain way,which greatly improves the acquisition efficiency.In order to obtain the same seismic records as the traditional source,we also need to separate the mixed records.In the process of seismic data acquisition,it will be affected by the surrounding environment(wind and grass,thunderstorms,insects and birds),so that the seismograph received by the seismometer will contain a variety of random noise interference,resulting in the reduction of seismic signal to noise ratio.Therefore,we need to remove random noise from seismic records to obtain seismic data with high signal-to-noise ratio and lay a foundation for subsequent processing.Sparse representation has been one of the most interesting topics in the field of digital signal analysis and processing in recent years.By means of a sparse representation method,the signal is transformed into a sparse domain,which enables us to obtain information that is difficult to obtain in time and space.By analyzing the sparse coefficient,setting the appropriate threshold to filter the smaller coefficient,separating the effective signal and the interference signal,and then transforming the signal in the sparse domain back to the time domain,the clean image after denoising can be obtained.This paper mainly studies the application of sparse representation in noise removal and mixed production noise separation of seismic data.In the principle of mixed acquisition,the basic concept and matrix representation of multi-source mixed acquisition are introduced,and three parameters related to mixed source acquisition are introduced,namely acquisition time ratio,source density ratio and mixing degree.In the process of formula derivation,the collection mode is classified by introducing mixed source operator.On this basis,the concept of pseudo-separation is introduced to make theoretical preparation for the separation of mixed data.In the sparse representation part,the important concepts such as sparsity and overcomplete dictionary are introduced,and the mathematical model of signal sparsity is introduced.This paper introduces two sparse optimization algorithms,orthogonal matching tracking algorithm and fast iterative threshold algorithm.The sparse algorithm is mainly introduced from two aspects: the fixed base dictionary and the learning dictionary.The fixed-base dictionary mainly introduces Fourier transform,discrete cosine transform,wavelet transform and curve transform.Learning dictionary mainly introduces two dictionary learning methods: principal component analysis and k-singular value decomposition.The mathematical derivation process of these algorithms is given,and the advantages and disadvantages of each algorithm are analyzed.In the seismic random noise removal part,this paper mainly studies the application of sparse representation method in seismic data to random noise removal.After sparse representation of seismic data containing noise,the corresponding sparse coefficient of the data is obtained.The purpose of denoising is realized by setting an appropriate threshold to remove the smaller coefficient.Based on the advantages of principal component analysis(PCA)algorithm and K-SVD dictionary learning algorithm,this paper proposes the method of combining PCA with K-SVD dictionary,and proves that the combined algorithm has better denoising effect and higher signal-to-noise ratio through experiments.In the separation of mixed data,this paper mainly studies the application of sparse representation in mixed mining separation.The pseudo-separated seismic data were transformed from the common shooting point domain to the common detection point domain,and the mixed production noise was removed by means of curve transform,K-SVD,principal component analysis,PCA combined curve transform and PCA combined K-SVD.By comparing and analyzing the separation effect and difference profile of several methods,it is proved that the method based on sparse representation is practical in mixed mining separation.The joint algorithm proposed in this paper has better separation effect and cleaner difference profile than the single sparse algorithm.