Automatic recognition of tidal bedforms from ground penetrating radar ({GPR}) data : building the training database

@inproceedings{bouche_automatic_RM2024,
 abstract = {Hydrogen plays a crucial role in the energy transition, yet its effective storage poses a significant challenge. Exploring storage possibilities near the surface, and elucidating the intricate mechanisms governing hydrogens gas leakage, necessitates extensive numerical modeling efforts. Furthermore, in the context of near-surface, Ground-Penetrating Radar (GPR) stands out as a remarkably efficient technique for characterization. This geophysical approach proves invaluable in identifying intricate geological formations, notably bedding and erosion surfaces. Enhancing this identification process could involve automation through artificial intelligence (AI). In this way, Convolutional Neural Network (CNN) are today widely used in the identification of anomalies and geological structures in geophysical data. Here we propose a method for generation of a synthetic data set to train a CNN for automatic recognition of unconformities in GPR data. Automated generation of synthetic geological models is developed with the Macro tool of SKUA-Gocad software, and corresponding synthetic GPR images are produced with GPRMax python library. A geological model of a geological reservoir composed of tidal bedforms structures, inspired by a pilot site for hydrogen storage in Saint-Emilion, in western France, will be used as an example.},
 author = {Bouche, Noemie and Caumon, Guillaume and Fratani, Amandine},
 booktitle = {Proc. 2024 RING Meeting},
 language = {en},
 pages = {15},
 publisher = {ASGA},
 title = {Automatic recognition of tidal bedforms from ground penetrating radar ({GPR}) data : building the training database},
 year = {2024}
}