Generative Adversarial Network for Seismic Data Interpolation

Summary Seismic data acquisition is the foundation of seismic exploration. When sampling at offset is too coarse during the acquisition, spatial aliasing will appear, affecting the accuracy of subsequent processing. In order to remove the spatial aliasing, the receiver spacing should be reduced, which can be achieved by interpolating one trace between every two traces. And the seismic data with spatial aliasing can be seen as regular missing data. Conditional generative adversarial networks (cGANs) are deep-learning models learning to generate new data with the same statistics as the training dataset based on the given input. In this abstract, a cGAN is designed for application to interpolation. To train the network, one geological model is created to synthesize seismic data. We use a synthetic dataset based on a new geological model and a field dataset to assess the performance of the trained network qualitatively and quantitatively. The test results indicate that the spatial aliasing can be removed effectively using the cGAN interpolation method.