Seismic stratigraphy interpretation via deep convolutional neural networks: A semi-supervised approa

To improving the performance of deep learning algorithms on seismic interpretation from a small amount of training data, this study presents an innovative workflow for seismic stratigraphy interpretation that consists with two components: (a) seismic feature self-learning (SFSL) and (b) stratigraphy model building (SMB), each of which is achieved in a deep CNN. While the latter is supervised and of the typical network architecture used in image segmentation, we design the former as unsupervised and requiring no knowledge from domain experts. Compared to the convolutional approaches, the proposed workflow makes it possible for the SMB network successfully inheriting the prior-knowledge for understanding the target seismic data from the SFSL one, so that the step of supervised learning can be efficiently completed by only a small amount of training data.

A real example (with 0.2% of the seismic data used for training)

References

Di, H., Z. Li, H. Maniar, and A. Abubakar, 2019, Seismic stratigraphy interpretation via deep convolutional neural networks: 89th SEG Technical Program Expanded Abstracts, 2358-2362. (Link) (PDF)