Deep Relative Geologic Time: A Deep Learning Method for Simultaneously Interpreting 3‐D Seismic Horizons and Faults

Extracting horizons and detecting faults in a seismic image are basic steps for structural interpretation and important for many seismic processing schemes. A common ground of the two tasks is to analyze seismic structures and they are related to each other. However, previously proposed methods deal...

Full description

Saved in:
Bibliographic Details
Published in:Journal of geophysical research. Solid earth Vol. 126; no. 9
Main Authors: Bi, Zhengfa, Wu, Xinming, Geng, Zhicheng, Li, Haishan
Format: Journal Article
Language:English
Published: Washington Blackwell Publishing Ltd 01-09-2021
Subjects:
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Extracting horizons and detecting faults in a seismic image are basic steps for structural interpretation and important for many seismic processing schemes. A common ground of the two tasks is to analyze seismic structures and they are related to each other. However, previously proposed methods deal with the tasks independently, and challenge remains in each of them. We propose a volume‐to‐volume neural network to estimate a relative geologic time (RGT) volume from a seismic volume, and this RGT volume is further used to simultaneously interpret horizons and faults. The network uses U‐shaped framework with attention mechanism to systematically aggregate multi‐scale information and automatically highlight informative features, and achieves high prediction accuracy with affordable computational costs. To train the network, we build thousands of 3‐D noisy synthetic seismic volumes and corresponding RGT volumes with realistic and various structures. We introduce a loss function based on structure similarity to capture spatial dependencies among seismic samples for better optimizing the network, and use multiple reasonable assessments to evaluate the predicted results. Trained by using synthetic data, our network outperforms the conventional approaches in recognizing structural features in field data examples. Once obtaining an RGT volume, we can not only obtain seismic horizons by simply extracting RGT constant surfaces but also detect faults that are indicated by lateral RGT discontinuities. To be able to deal with large seismic volumes, we further propose a workflow to first estimate sub‐volumes of RGT and merge them to obtain a full RGT volume without boundary artifacts. Plain Language Summary We propose a volume‐to‐volume deep network to accurately compute an RGT volume from the seismic image volume without any manual picking constraints, which is further used to simultaneously interpret horizons and faults. This network is simplified from the originally more complicated UNet and supplemented by multi‐scale residual learning and attention mechanisms to achieve acceptable computational costs but still preserve the high prediction accuracy for RGT estimation. We use a non‐trivial criterion based on structural similarity to optimize the network parameters. We train the network by using synthetic data, which are automatically generated by adding folding and faulting structures to an initial flat seismic and RGT volumes. Our CNN model not only shows promising prediction performance on the synthetic data excluded in training but also works well in the multiple field seismic volumes that are recorded at totally different surveys to reliably capture complex structural features such as crossing faults and complicatedly folded horizons. The applications in the three field data examples demonstrate the prediction performance and reliable generalization ability of the trained CNN model. These results can provide valuable information by revealing complex subsurface structures, which is of importance to many human activities ranging from natural source exploration to geothermal energy production. Key Points We propose an encoder‐decoder network with attention mechanism to estimate relative geologic time (RGT) volumes from 3D seismic images We train the network with a criterion of structural similarity which enables the network capture seismic structural interdependences Our method can deal with large 3D seismic images and estimate RGT volumes from which all horizons and faults can be automatically extracted
ISSN:2169-9313
2169-9356
DOI:10.1029/2021JB021882