Automated interpretation of top and base salt using deep-convolutional networks

Abstract

We present a new automated workflow based on machine learning that can significantly reduce the amount of manual interpretation of top and base salt boundaries. Manual interpretation of salt boundaries on large seismic surveys with complex salt geometry is a time-consuming task. The interpreters manually pick a large number of surface control points line-by-line through the seismic volume. In this new method, we will replace the manual surface control picking with automation by using a machine learning approach. In the first part of the workflow we use a convolutional neural network that is designed to detect the top-of-salt boundary. Subsequently, a second convolutional network is then designed to detect the base of salt boundary. In both cases, the training data are picked as 2D subsections together with the corresponding manual interpretations in a specific seismic survey. The two trained networks are then evaluated both on the seismic data used in the training and on seismic data not used in the training. In both cases, we produce a top- and base-salt interpretation that coincides with the major parts of the manual interpretations. This new automated workflow has the potential to reduce the interpretation turnaround time of both top and base of salt. Presentation Date: Tuesday, October 16, 2018 Start Time: 8:30:00 AM Location: 204B (Anaheim Convention Center) Presentation Type: Oral