Membership renewals open for 2024 - Click here

Deep learning for seismic interpretation

Event Type

Event Date

Tuesday, April 14, 2020

Event Location

Event Address


Event Start

10 AM (Beijing)

Event End

11 AM (Beijing)

Event Details


Format: Virtual Webinar. 45 min. presentation followed by 15 min. Q&A

Please note that two sessions will be given at different dates listed below.

Session 1, Tuesday, April 14, 2020, 10 am to 11 am Beijing time Register Here

Session 2, Thursday, April 16, 2020, 9 am to 10 am IST time (India) Register Here



Seismic interpretation involves detecting and extracting structural information, stratigraphic features, and geobodies from seismic images. Although numerous automatic methods have been proposed, seismic interpretation today remains a highly time-consuming task which still requires significant human efforts. The conventional seismic interpretation methods or workflows are not automated or intelligent enough to efficiently or accurately interpret the rapidly increasing seismic data sets, which leaves significantly more data uninterpreted than interpreted.

We improve automatic seismic interpretation by using CNNs (convolutional neural networks) which recently have shown the best performance in detecting and extracting useful image features and objects. One main limitation of applying CNNs in seismic interpretation is the preparation of many training data sets and especially the corresponding geologic labels. Manually labeling geologic features in a seismic image is highly time-consuming and subjective, which often results in incompletely or inaccurately labeled training images. To solve this problem, we propose a workflow to automatically build diverse geologic models with geologically realistic features. Based on these models with known geologic information, we further automatically create numerous synthetic seismic images and the corresponding ground truth of geologic labels to train CNNs for geologic interpretation in field seismic images. Accurate interpretation results in multiple field seismic images show that the proposed workflow simulates realistic and generalized geologic models from which the CNNs effectively learn to recognize real geologic features in field images.

In this lecture, I would like the share you with our research experience on the following topics:

Automatic preparation of training data sets and labels;

CNN for fault detection, fault orientation estimation, and fault surface construction;

CNN for relative geologic time and seismic horizons;

CNN for seismic geobody tracking;

CNN-based multitask learning in seismic interpretation.