Carissimi Associati SGI,
Carissimi Soci Società Associate,
Carissimi Soci Società Associate,
su indicazione del Prof. Domenico Liotta (Università di Bari), vi segnaliamo il seminario online del Prof. Cédric John (Queen Mary University of London, https://www.qmul.ac.uk/deri/deri-people/deri-staff-/profile/cedric-m-john.html) dal titolo "Sharper, Faster, Smarter: How Computer Vision could help Geothermal Characterisation in Carbonates", che si terrà il 6 novembre 2025 alle ore 13.00.
Link Teams per il collegamento
Per ulteriori dettagli vi invitiamo a leggere l'abstract riportato di seguito.
Cordiali saluti,
La Segreteria SGI
Abstract
Carbonate rocks exhibit heterogeneities from the seismic scale to the pore level, which pose challenges in interpretation, especially in mixed carbonate-clastic systems. Aside from the well-known 'scale gap' issue where heterogeneities change depending on the scale of observation, humans also lack consistency in how they interpretate observational data. Accurate interpretation of carbonate grains and textures is essential for deducing depositional environments and stratigraphic sequences. Traditional manual interpretation of geological data is labour-intensive and often prone to inaccuracies, potentially leading to flawed subsurface models.
In this talk, I will present over eight years of research from my team on using computer vision to enhance the interpretation of geological data in these complex systems, focusing on core images, logs, and seismic data, mostly coming from ODP Leg 194 (a Miocene mixed carbonate-clastic system).
Our research has shown that convolutional neural networks (CNNs) can interpret core data with greater speed and accuracy than experienced geologists. This approach is cost-effective and biases from the models can be mitigated through visual inspection of edge cases and the consistency of bias across different wells and basins. Transfer learning has been pivotal, along with trials in individual grain classification.
Carbonate rocks exhibit heterogeneities from the seismic scale to the pore level, which pose challenges in interpretation, especially in mixed carbonate-clastic systems. Aside from the well-known 'scale gap' issue where heterogeneities change depending on the scale of observation, humans also lack consistency in how they interpretate observational data. Accurate interpretation of carbonate grains and textures is essential for deducing depositional environments and stratigraphic sequences. Traditional manual interpretation of geological data is labour-intensive and often prone to inaccuracies, potentially leading to flawed subsurface models.
In this talk, I will present over eight years of research from my team on using computer vision to enhance the interpretation of geological data in these complex systems, focusing on core images, logs, and seismic data, mostly coming from ODP Leg 194 (a Miocene mixed carbonate-clastic system).
Our research has shown that convolutional neural networks (CNNs) can interpret core data with greater speed and accuracy than experienced geologists. This approach is cost-effective and biases from the models can be mitigated through visual inspection of edge cases and the consistency of bias across different wells and basins. Transfer learning has been pivotal, along with trials in individual grain classification.