Oriented ellipsoidal DBSCAN for clustering faults from deep learning attributes
Samuel Chambers, Jesse Lomask · 2024
This abstract introduces a new method, Oriented Ellipsoidal Density-Based Clustering and Noise (OE-DBSCAN), for clustering fault attributes obtained from deep learning techniques in seismic data analysis. The goal is to accurately cluster faults, even in complex scenarios like intersecting or curved faults and gaps due to data anomalies. By transforming attribute volumes into a structured dataset of spatial coordinates and orientation attributes, OE-DBSCAN uses an orientable ellipsoid to improve clustering accuracy along fault strike and dip. Comparative testing with traditional DBSCAN shows superior performance, especially in handling gaps in fault predictions.