We describe SaltSeg: a high capacity deep convolutional neural network (CNN) architecture that achieves human level interpretation accuracy on seismic images. SaltSeg is primarily used for salt interpretation and is a key component of a deep learning based fully automated salt model building pipeline. It is designed to work on low resolution, noisy, incorrectly migrated seismic images as is typically encountered during the model building stage and achieves human level interpretation accuracy on such images and can be easily modified for other geologic interpretation tasks. We give an indepth description of the key building blocks of SaltSeg and describe a novel integration of a β-variational autoencoder (VAE) branch with a standard encoder-decoder network that leads to significant boost in interpretation accuracy. We validate our results using real data images from surveys in the Gulf of Mexico.

Presentation Date: Tuesday, September 17, 2019

Session Start Time: 9:20 AM

Presentation Time: 10:35 AM

Location: Poster Station 2

Presentation Type: Poster

This content is only available via PDF.
You can access this article if you purchase or spend a download.