A weakly supervised method for improving the resolution of seismic data based on cycle generative adversarial network

Abstract

In order to solve the shortcomings of traditional methods in improving the resolution of seismic data, and the problem that the strongly supervised methods rely too much on labels, in this paper, we proposed a weakly supervised method to improve the resolution of 3D seismic data by using a cycle generative adversarial network, the label is a high-resolution seismic data from another work area that is not paired with low resolution data. The network learns a distribution from the low-resolution seismic data to the high-resolution seismic data, and generates a high-resolution result similar to the high-resolution label. That is, by learning two unrelated high-and low-resolution data to achieve the weakly supervised learning process. Experiments prove that the weakly supervised method can obtain similar result as the strongly supervised method.