Real-time seismic image interpretation via deconvolutional neural network

To address the limitation of low efficiency for classifying seismic features from large datasets, this study aims at implementing the deconvolutional neural network (DCNN) for the purpose of real-time seismic interpretation, so that all the important features in a seismic image can be identified and interpreted both accurately and simultaneously. The performance of the new DCNN tool is verified through application of segmenting the F3 seismic dataset into nine major features, including salt domes, strong reflections, steep dips, etc. Good match is observed between the results and the original seismic signals, indicating not only the capability of the proposed DCNN network in seismic image analysis but also its great potentials for real-time seismic feature interpretation of an entire volume.

Result analysis

The 3D view of the feature volume generated by the proposed DCNN. It is also clipped to 6 randomly-selected sections that were not used in training. Note the correct labelling of the major features, such as salt domes as well as the overlaying strong continuous reflection in the bottom and the steep dips in the middle of important depositional implications.

Conclusions

We have implemented the deconvolutional neural network (DCNN) into assisting seismic interpretation. The major superiority of such tool lies that the DCNN directly works on a seismic section and thereby is capable of label all the features in it at real time, which is more efficient than the traditional interpretation tools and the classification techniques (e.g., convolutional neural network). The success of the proposed DCNN on labelling the F3 block into 9 classes indicates its great potentials for assisting 3D seismic interpretation in a both fast and accurate manner.

Acknowledgments

We thank Peter Bormann as well as his colleagues at ConocoPhillips for labelling and sharing the F3 block seismic dataset. The deconvolutional neural network algorithm is implemented based on the open-source Python package TensorFlow developed by Google Brain.

References:

1. Di, H., Z. Wang, and G, AlRegib, 2018, Real-time seismic image interpretation via deconvolutional neural network: 88th Annual SEG Meeting Extended Abstracts, 2051-2055.