Suppressing spatial dispersion of seismic finite-difference modeling with the improved pix2pix algorithm

Abstract

Seismic finite-difference (FD) modeling inevitably suffers from the spatial dispersion artifacts. In this paper, we propose a new method to suppress the spatial dispersion of seismic FD modeling using the improved pix2pix algorithm by introducing the Sobel operator. The new method involves the following steps. First, a training data set which consists of a small number of wavefield data is generated. The training data set is composed of the high-accuracy wavefield data modeled by the high-order FD method and the low-accuracy wavefield data modeled by the low-order FD method. Second, the entire set of the wavefield data polluted by the spatial dispersion is computed using the low-order FD method. Third, the improved pix2pix neural network is trained and applied to the entire wavefield data set to suppress the spatial dispersion. Numerical tests on different modeling examples demonstrate the effectiveness of the proposed method, which includes the great accuracy in suppressing the spatial dispersion and the high efficiency for the large scale FD modeling.