Penda Biondokin, Eric; Bavandsavadkoohi, Mojtaba; Tirdad, Shiva et Gloaguen, Erwan 
ORCID: https://orcid.org/0000-0002-9400-0276
(2025).
Training Super-Resolution deep learning algorithms for high resolution aeromagnetic maps generation from low resolution aeromagnetic maps.
In: European Geosciences Union (EGU) General Assembly, 27 avril-2 mai 2025, Vienne, Autriche.
(Soumis)
Résumé
The province of Quebec (Canada) is regarded as the principal mining Province of Canada due to its substantial exploitable reserves and the significant contribution of its mineral production to the national GDP. Nevertheless, Vast areas, such as northern Quebec, remain insufficiently covered in terms of geoscientific data, limiting the understanding of their mineral exploration potential.
Aeromagnetic data are widely employed for large-scale reconnaissance to map geological structures and guide geologists in identifying exploration targets or defining new prospects. However, the only data that covers the entire area are low-resolution aeromagnetic data, with high-resolution datasets being sporadically available. This low resolution restricts the interpretability of regional data, as certain geological structures remain hidden by coarse sampling intervals. To enhance geological mapping, it is imperative to improve the resolution of aeromagnetic data to reveal structures such as faults, lineaments, and lithological boundaries that are otherwise undetectable in low-resolution geophysical signatures. While acquiring high-resolution data is an ideal solution, the high costs and vast territorial coverage required render this approach challenging in the short term. As an alternative, the advent of artificial intelligence (AI), particularly deep learning, offers promising avenues for exploration. In this study, we adapted and retrained 4 super-resolution deep learning algorithms to generate high resolution aeromagnetic maps from low resolution ones. To avoid bias due to spatial correlation, we split the data sets into a training set covering the southern part of Québec and validation being the Northern part. Each of the AI codes were trained on the same datasets leading to optimal hyperparameters for each algorithm. The AI-generated results for all the 4 algorithms successfully reconstruct high-resolution regional aeromagnetic maps in the training sets compared to measured high resolution data providing reliable high resolution maps for geological mapping. Finally, we generated four high resolution aeromagnetic maps for entire Province including the northern part. This innovative approach holds the potential to revolutionize geophysical exploration, facilitating the discovery of untapped natural resources in underexplored areas.
| Type de document: | Document issu d'une conférence ou d'un atelier |
|---|---|
| Mots-clés libres: | -- |
| Centre: | Centre Eau Terre Environnement |
| Date de dépôt: | 26 mars 2025 18:25 |
| Dernière modification: | 26 mars 2025 18:25 |
| URI: | https://espace.inrs.ca/id/eprint/16403 |
Gestion Actions (Identification requise)
 |
Modifier la notice |
