Deep-learning-based global feature capture for seismic data reconstruction

What is it about?

Big gaps, or consecutive missing traces, severely reduces the quality of seismic data. Though existing CNN-based interpolation methods perform well on reconstructing regularly or irregularly missing traces, they are not powerful enough to recover such big gaps because they can only “see” and utilize local information. We turn to Transformer that can integrate global information, and conduct extensive experiments to explore its interpolation capability of reconstructing the consecutive missing cases in seismic data.

Featured Image