Full text loading...
-
Seismic classification: A Thalweg tracking/machine learning approach
- Source: First Break, Volume 39, Issue 3, Mar 2021, p. 59 - 64
-
- 01 Mar 2021
Abstract
We describe a machine learning seismic classification workflow in which the thousands of class labels needed for training a deep graph are automatically generated from just a handful of manually picked seed positions. The class labels are generated by a Thalweg tracker. This special kind of connectivity filter grows a 3D body of user-defined size from a single seed position by adding only one point at a time. The user controls the size such that the tracker stays within one seismic class. The shape of the growing body is the main criterion for deciding when and where the tracker starts tracking another class. We present two examples. The first example is a 3D seismic facies classification of a setting with stacked meandering channels. We classify the target interval into eight seismic facies classes. In the second example, we extract turbidite channels from a two-pass gradient (i.e., second derivative) attribute volume.