Salt interpretation on seismic data has historically been a very manual process, requiring weeks or even months to complete on one 3D seismic survey. The accuracy of the interpreted salt boundary is critical for sub-salt imaging and subsequent drilling for oil and gas. The nature of the salt problem can be reduced to a binary classification problem that is well suited to modern machine learning (ML) algorithms: each location on an image either contains salt or sediment. Seismic surveys are collected and processed in different ways, which poses a challenge to traditional ML methods that rely on statistical similarity between training data and prediction data, especially where limited training data are available. We propose to use a supervised ML approach that treats each seismic survey independently. In particular, we show that an adaptive U-Net approach yields accurate salt bodies in minutes rather than weeks and requires minimal training data.
Presentation Date: Monday, October 12, 2020
Session Start Time: 1:50 PM
Presentation Time: 2:15 PM
Location: 351F
Presentation Type: Oral
