Research On High Dimensional Seismic Signal Processing Method Based On Tensor Sparse Coding

Due to the influence of field acquisition,there are problems with noise pollution and insufficient sampling rate of the seismic signals,which will affect the interpretation of seismic data.Therefore,it is very important to process the raw seismic records collected in the field.This can suppress noise interference and improve the signal-to-noise ratio,resolution,and fidelity of seismic data.Tensor sparse coding is a multi-dimensional data representation method that can make good use of redundant information between multi-dimensional seismic data,and has powerful signal processing and feature extraction capabilities.This thesis combines the theory of tensor sparse coding with compressed sensing data reconstruction and dictionary learning,and proposes reasonable and effective algorithms from seismic track interpolation and seismic data denoising.The specific work is summarized as follows:Firstly,to solve the problem of spatial aliasing caused by insufficient spatial sampling rate of seismic data,this thesis combines tensor sparse coding with compressed sensing data reconstruction theory,and proposes a seismic trace interpolation method based on tensor joint sparse coding.This method introduces tensor product to extend two-dimensional dictionary learning to three-dimension.In this method,the tensor sparse coefficients and the tensor dictionary are alternately iteratively solved.The tensor-based iterative contraction threshold algorithm is used to process the tensor sparse coefficients,and the tensor dictionary is solved using the Lagrangian dual method to increase the calculation rate.This method uses three-dimensional dictionary learning to make more efficient use of the redundant information of seismic data,realize seismic trace interpolation,and improve the resolution of seismic data.Secondly,for the problem of seismic data being polluted by noise,this thesis proposes a seismic data denoising method based on tensor coherence constraint dictionary learning.This method effectively combines coherent constrained denoising with tensor sparse coding by defining tensor coherence ratios.The method solves the problem that the conventional dictionary learning denoising method relies on the noise variance prior and the effect is not good when the noise variance changes.In theprocess of alternate iterations,the tensor coherent matching pursuit algorithm is used to solve the tensor sparse coefficients.At the same time,the singular value decomposition of tensor decomposition is introduced to extend the K-SVD dictionary learning algorithm to three dimensions to obtain the K-TSVD algorithm for solving the tensor dictionary,thereby achieving the tensor coherence constrained denoising of seismic data.In this thesis,the above two methods have been experimentally verified in the theoretical model and the three-dimensional seismic data of the actual work area and compared with the traditional methods.It is found that this method has certain advantages compared with the traditional methods and can achieve better practical results.