The gradient preconditioning approach based on seismic wave energy can effectively avoid the huge memory consumption of the gradient preconditioning algorithms based on the Hessian matrix. However, the accuracy of thi...The gradient preconditioning approach based on seismic wave energy can effectively avoid the huge memory consumption of the gradient preconditioning algorithms based on the Hessian matrix. However, the accuracy of this approach is prone to be influ- enced by the energy of reflected waves. To tackle this problem, the paper proposes a new gradient preconditioning method based on the energy of transmitted waves. The approach scales the gradient through a precondition factor, which is calculated by the ‘ap- proximate transmission wavefield’ simulation based on the nonreflecting acoustic wave equation. The method requires no computing nor storing of the Hessian matrix and its inverse matrix. Furthermore, the proposed method can effectively eliminate the effects of geometric spreading and disproportionality in the gradient illumination. The results of model experiments show that the time-domain full waveform inversion (FWI) using the gradient preconditioning based on transmitted wave energy can achieve higher inversion accuracy for deep high-velocity bodies and their underlying strata in comparison with the one using the gradient preconditioning based on seismic wave energy. The field marine seismic data test shows that our proposed method is also highly applicable to the FWI of field marine seismic data.展开更多
Low-frequency band-shaped swell noise with strong amplitude is common in marine seismic data.The conventional high-pass fi ltering algorithm widely used to suppress swell noise often results in serious damage of effec...Low-frequency band-shaped swell noise with strong amplitude is common in marine seismic data.The conventional high-pass fi ltering algorithm widely used to suppress swell noise often results in serious damage of effective information.This paper introduces the residual learning strategy of denoising convolutional neural network(DnCNN)into a U-shaped convolutional neural network(U-Net)to develop a new U-Net with more generalization,which can eliminate low-frequency swell noise with high precision.The results of both model date tests and real data processing show that the new U-Net is capable of effi cient learning and high-precision noise removal,and can avoid the overfi tting problem which is very common in conventional neural network methods.This new U-Net can also be generalized to some extent and can eff ectively preserve low-frequency eff ective information.Compared with the conventional high-pass fi ltering method commonly used in the industry,the new U-Net can eliminate low-frequency swell noise with higher precision while eff ectively preserving low-frequency eff ective information,which is of great signifi cance for subsequent processing such as amplitude-preserving imaging and full waveform inversion.展开更多
基金support of the NSFCShandong Joint Fund for Marine Science Research Centers (No. U1606401)the National Natural Science Foundation of China (Nos. 41574105 and 41704114)+1 种基金the National Science and Technology Major Project of China (No.2016ZX05027-002)Taishan Scholar Project Funding (No. tspd20161007)
文摘The gradient preconditioning approach based on seismic wave energy can effectively avoid the huge memory consumption of the gradient preconditioning algorithms based on the Hessian matrix. However, the accuracy of this approach is prone to be influ- enced by the energy of reflected waves. To tackle this problem, the paper proposes a new gradient preconditioning method based on the energy of transmitted waves. The approach scales the gradient through a precondition factor, which is calculated by the ‘ap- proximate transmission wavefield’ simulation based on the nonreflecting acoustic wave equation. The method requires no computing nor storing of the Hessian matrix and its inverse matrix. Furthermore, the proposed method can effectively eliminate the effects of geometric spreading and disproportionality in the gradient illumination. The results of model experiments show that the time-domain full waveform inversion (FWI) using the gradient preconditioning based on transmitted wave energy can achieve higher inversion accuracy for deep high-velocity bodies and their underlying strata in comparison with the one using the gradient preconditioning based on seismic wave energy. The field marine seismic data test shows that our proposed method is also highly applicable to the FWI of field marine seismic data.
基金the Key R&D project of Shandong Province(No.2019JZZY010803)the Central Universities(No.201964016),the National Natural Science Foundation of China(No.41704114)+2 种基金the National Science and Technology Major Project of China(No.2016ZX05027-002)Taishan Scholar Project Funding(No.tspd20161007)the China Scholarship Council(No.201906335010).
文摘Low-frequency band-shaped swell noise with strong amplitude is common in marine seismic data.The conventional high-pass fi ltering algorithm widely used to suppress swell noise often results in serious damage of effective information.This paper introduces the residual learning strategy of denoising convolutional neural network(DnCNN)into a U-shaped convolutional neural network(U-Net)to develop a new U-Net with more generalization,which can eliminate low-frequency swell noise with high precision.The results of both model date tests and real data processing show that the new U-Net is capable of effi cient learning and high-precision noise removal,and can avoid the overfi tting problem which is very common in conventional neural network methods.This new U-Net can also be generalized to some extent and can eff ectively preserve low-frequency eff ective information.Compared with the conventional high-pass fi ltering method commonly used in the industry,the new U-Net can eliminate low-frequency swell noise with higher precision while eff ectively preserving low-frequency eff ective information,which is of great signifi cance for subsequent processing such as amplitude-preserving imaging and full waveform inversion.
