Generative adversarial networks in seismic data processing (Stephen Alwon, Schlumberger)

Generative adversarial networks (GANs) are a class of machine learning techniques that involve two networks trained simultaneously to generate a desired outcome. These schemes have had success in many traditional image processing tasks, such as style transfer and super-resolution, but are relatively unexplored in geophysics. We outline the underlying theory behind GANs and present networks that can perform traditional seismic processing tasks such as noise attenuation and trace interpolation.

Convolutional neural network for Salt model building in the Gulf of Mexico (Sribharath Kainkaryam, TGS )

Salt model building is one of the most time consuming, labor intensive and difficult to automate processes in the entire depth imaging workflow requiring significant human effort. The challenge and need for automating salt interpretation is well recognized by the seismic imaging community and has seen many applications of deep learning based convolutional neural network (CNN) architectures to carry out this task. However, significant challenges remain for reliable production-scale deployment of CNN based methods for salt model building owing to poor generalization capabilities of these networks. When used on new surveys, never seen by the CNN models during the training stage, the interpretation accuracy of these models drops significantly. To remediate this key problem, we introduce a U-shaped encoder-decoder type CNN architecture and use a specialized regularization strategy aimed at reducing the generalization error of the network. Our regularization scheme perturbs the ground truth labels in the training set. Two different perturbations are discussed: one which randomly changes the labels of the training set, flipping salt labels to sediments and vice versa and the second which smooths the labels.  We demonstrate that such perturbations act as a strong regularizer preventing the network from making low entropy, highly confident predictions on the training set and thus reducing overfitting. An ensemble strategy is also devised for test time augmentation that is shown to further improve the accuracy. The robustness, in terms of vastly improved generalization capability as well as improved interpretation accuracy of our training strategy is demonstrated with real data examples from the Gulf of Mexico. 

Sponsors: