| As one of the important means of seismic data processing,Radon inversion provides premise and guarantee for seismic data interpretation.To be specific,Radon transform faces two main problems.Firstly,the calculation efficiency is low.Complex and large-scale seismic data inversion involves the operation of large operator matrix,which takes a long time and costs a lot.Secondly,the resolution is difficult to meet the requirements.Under working conditions,low sampling of seismic data(limited aperture)will inevitably lead to low-resolution inversion results,which can only be remedied by improving the algorithms.In this paper,based on the traditional iterative inversion algorithms,the complex nonlinear characterization ability of neural network is utilized to speed up the inversion and improve the stability of the solution,so as to realize the high-resolution sparse representation of time-varying Radon domain.A method called fast sparse hyperbolic Radon transform(FSHRT)is designed.The specific content is summarized as follows:(1)Considering the characteristics of seismic data,the properties of time-varying Radon transform are studied on the basis of the theory of solving underdetermined inverse problem.Based on the inversion model,the deconvolution mapping relationship between the inversion conjugate solution and the sparse Radon coefficients is established.And the inversion problem is transformed into the image deblurring problem.(2)Considering the complex nonlinear characteristics of deconvolution problem,based on deep learning theory,sparse prior and regularization theory,a data-driven deconvolution method is established.The deconvolution neural network is designed to complete the tasks of Radon coefficients error elimination and resolution improvement,and the structure and parameters of the neural network are optimized.(3)On the premise of realizing fast and effective high-resolution inversion process,the intrinsic mechanism of sparse representation by deconvolution model is deeply studied.The synthetic data and marine data are used to verify the above theoretical methods.After comparing with the conventional high-resolution algorithms,the errors and its sources are analyzed,and finally the method is optimized.The FSHRT is applied to multiple suppression.The processing results of synthetic seismic data and marine seismic data confirm the efficiency and reliability of the proposed method. |
PDF Full Text Request