Today, in our series on Machine Learning Workflows, we go back to the origins of the OpendTect Machine Learning platform. OpendTect started life as a neural network-based seismic pattern recognition and attribute processing system. The primary goal of the original system was to create Chimney Cubes for fluid migration path interpretation. The software was used for geohazard interpretation and for de-risking hydrocarbon charge and seal problems.
The salt dome body is extracted with a 2D-Unet using the workflow described in an earlier post in this series. The anomalies are visualized with the energy attribute in a volume viewer.
The horizon above the shallow anomaly is an RGB blend of 3 spectral components. The alpha channel shows the RMS of the Thinned Fault Likelihood attribute. The Chimney Cube is created with a supervised learning approach. A fully connected Multi-Layer Perceptron is trained on attributes extracted at user-picked locations of Chimneys and Not-Chimneys. The video shows how this is done.