Mapping Geology using Textural Feature Extraction and Unsupervised Community Detection Models on Airborne Geophysics
Abstract:
Airborne geophysics can provide useful information that can assist in large scale geological mapping. However, this data can be hard to interpret, especially when there is limited ground truth due to varying cover thickness. Here, we demonstrate how computer vision based feature extraction combined with unsupervised community detection models can be used to identify areas with similar geological textures to aid in interpretation and mapping.
In this study, a pre-trained convolutional neural network (CNN) model was used to extract textural information from the airborne geophysics. A uniform manifold approximation and projection (UMAP) was fitted to the data to reduce the dimensionality and graph the relationships.
Community detection algorithms can use the graph produced by the UMAP to detect groups with similar properties. We tested the Louvain (using both Potts and Dugue modularities), Leiden and Walktrap algorithms using gravity and total magnetic intensity (TMI) data from an area of the Gawler Craton in South Australia. The results were compared to the mapped Archean to Early Mesoproterozoic geology to assess model performance.
Dr Katie Silversides:
Katie Silversides is a data scientist and geologist at Datarock. She works in the applied science team, solving mining and exploration problems using a combination of domain expertise and advanced machine-learning techniques.
Katie completed her PhD in Geology at the University of Sydney. She then worked at the Rio Tinto Centre for Mine Automation, working on applying machine learning to solve different problems relating to orebody geology. This was followed by a position in DARE working on geology and hydrology projects, focusing on missing data, uncertainty and model selection.
Негізгі бет Mapping Geology using Textural Feature Extraction and Unsupervised Community Detection Models
Пікірлер