Banca de DEFESA: MATHEUS FELIPE FREITAS TOMAZ
Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.STUDENT : MATHEUS FELIPE FREITAS TOMAZ
DATE: 26/09/2025
TIME: 14:00
LOCAL: Plataforma Google Meet
TITLE:
Reconstruction of the wavefield in multicomponent OBN experiments using generative neural networks based on supervised learning
KEY WORDS:
Deep Learning, Pix2Pix, Denoising Diffusion Probabilistic Models, OBN wavefield
PAGES: 108
BIG AREA: Ciências Exatas e da Terra
AREA: Física
SUMMARY:
The propagation of artificially generated waves is one of the main tool of exploration geophysics and is widely employed in the oil and gas industry for subsurface geological mapping. In particular, ocean-bottom node (OBN) experiments with multicomponent sensors allow for the complete acquisition of the wavefield through measurements of pressure (hydrophone) and the three components of particle velocity (geophone). Despite the richness of the information, deploying such nodes involves high costs and operational challenges. In this context, we investigate the use of two generative models based on supervised deep learning - the Pix2Pix and the Denoising Diffusion Probabilistic Model (DDPM) - to convert signals acquired by hydrophones into equivalent records captured by geophones. These models can learn the direct mapping between pressure data (input) and velocity components (output), leveraging the kinematic and dynamic properties of the wavefield without the need to impose explicit physical constraints. Using a 2D scenario and a synthetic acoustic dataset, we evaluate the performance of the models under three sparsity conditions, highlighting their characteristics, advantages, and limitations. The results demonstrate the models’ ability to satisfactorily reproduce the physical features of the wavefield. Quantitative metrics, such as the relative root mean square error (RMSE) and the structural similarity index measure (SSIM), corroborate the quality of the predictions, while analyses in the migrated data domain suggest that these models may represent a promising alternative to partially replace multicomponent nodes with hydrophones.
COMMITTEE MEMBERS:
Externo à Instituição - EDWIN HUMBERTO FAGUA DUARTE - UFRN
Presidente - 2492756 - JOAO MEDEIROS DE ARAUJO
Externo à Instituição - JORGE LUIS LOPEZ
Interno - 2411793 - LEONARDO DANTAS MACHADO
Externo à Instituição - Pedro da Silva Peixoto - USP