Better Seismic Facies Analysis - Imposing seismic interpretation constraints on AI

Research papers
Applying Deep Learning algorithms to solve some geophysical challenges and gaining the prototype (ground-zero) is reasonably straightforward. You take an open dataset with labels(or create your own), download a state-of-the-art neural network, allocate some compute on Google Cloud, grab a recent publication and coffee, and you are ready to rock. However, it is always challenging to move from just providing prediction to a helpful tool, following existing interpretation rules.

The paper "Haibin Di et al., 2021, Imposing interpretational constraints on a seismic interpretation convolutional neural network." approaches the challenge by providing additional information for neural networks during the training process. Instead of relying only on raw seismic amplitude information, the authors include an RGT image as an input for the neural network that provides a better prediction of seismic facies. For those unfamiliar with RGT (Relative Geological Time) - it is a color-coded continuous value that provides information about formation time from young to old. Moreover, the paper short-listed a set of objective and subjective constraints commonly used by interpreters and utilized them to guide CNN.

The link for the post on LinkedIn
The link for the paper