Creating horizon interval class labels in OpendTect If your goal is to develop a workflow for horizon tracking based on Machine Learning, you may consider a two-step approach: 1...
Creating horizon interval class labels in OpendTect
If your goal is to develop a workflow for horizon tracking based on Machine Learning, you may consider a two-step approach: 1) image-to-image segmentation followed by 2) extracting the boundaries of the segmented volume as horizons.
In image-to-image segmentation, a model transforms a seismic input of certain dimensions into a segmented output of equal dimensions. Seismic segmentation is supported in OpendTect ML with Keras-based 2D - and 3D Unets. The user controls the dimensions of the input shapes. (Researchers are of course free to change the architecture of these models, or to develop their own models inside the OpendTect / Python ecosystem).
To create training examples for this workflow, you need two volumes: 1) the seismic input cube and 2) the target volume with class labels. The latter volume is created via OpendTect’s attribute engine.
The free Delft survey (https://lnkd.in/d5NfmeZ) on TerraNubis features an attribute set called “Create_class_labels” to do just that (see slider).
As of version 6.6.4, OpendTect will not check for license keys in this project. This means Delft can be used to develop and test OpendTect Machine Learning applications without a license for OpendTect Pro and Machine Learning.
