The resolution achievable by standard seismic imaging techniques is limited in theory by the physics of subsurface wave propagation, and in practice by the computational expense associated with imaging at high frequencies. Modern machine learning approaches, and in particular an emerging class of "deep" learning algorithms, offer opportunities to address both limitations by establishing relationships between low- and high-resolution versions of seismic images. Here, a generative adversarial network (GAN) is trained to produce higher-resolution realizations of previously-unseen low-resolution input images. Early results on both 2D and 3D synthetic data show promise for this method as a means to enhancing seismic image quality and improving imaging and interpretation workflows.
Presentation Date: Wednesday, October 17, 2018
Start Time: 1:50:00 PM
Location: 204B (Anaheim Convention Center)
Presentation Type: Oral
