2021/10/27 by Matthieu Cedou, Erwan Gloaguen, Matthieu, Cedou +9
Computer Science · Earth and Planetary Sciences · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #FOS: Physical sciences #Geochemistry and Geologic Mapping #Geological Modeling and Analysis #Geophysical and Geoelectrical Methods #Geophysics (physics.geo-ph) #Image Processing and 3D Reconstruction
paper · pdf · doi:10.48550/arxiv.2110.14440
openalex publication_date 2021/10/27 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Airborne magnetic data are commonly used to produce preliminary geological\nmaps. Machine learning has the potential to partly fulfill this task rapidly\nand objectively, as geological mapping is comparable to a semantic segmentation\nproblem. Because this method requires a high-quality dataset, we developed a\ndata augmentation workflow that uses a 3D geological and magnetic\nsusceptibility model as input. The workflow uses soft-constrained Multi-Point\nStatistics, to create many synthetic 3D geological models, and Sequential\nGaussian Simulation algorithms, to populate the models with the appropriate\nmagnetic distribution. Then, forward modeling is used to compute the airborne\nmagnetic responses of the synthetic models, which are associated with their\ncounterpart surficial lithologies. A Gated Shape Convolutional Neural Network\nalgorithm was trained on a generated synthetic dataset to perform geological\nmapping of airborne magnetic data and detect lithological contacts. The\nalgorithm also provides attention maps highlighting the structures at different\nscales, and clustering was applied to its high-level features to do a\nsemi-supervised segmentation of the area. The validation conducted on a portion\nof the synthetic dataset and data from adjacent areas shows that the\nmethodology is suitable to segment the surficial geology using airborne\nmagnetic data. Especially, the clustering shows a good segmentation of the\nmagnetic anomalies into a pertinent geological map. Moreover, the first\nattention map isolates the structures at low scales and shows a pertinent\nrepresentation of the original data. Thus, our method can be used to produce\npreliminary geological maps of good quality and new representations of any area\nwhere a geological and petrophysical 3D model exists, or in areas sharing the\nsame geological context, using airborne magnetic data only.\n