Banca de QUALIFICAÇÃO: THIAGO THEIRY DE OLIVEIRA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : THIAGO THEIRY DE OLIVEIRA
DATE: 23/02/2026
TIME: 09:00
LOCAL: Sala Virtual
TITLE:

Automated mapping of the boundary between continental and oceanic crusts (COB/COT) based on Machine Learning.

KEY WORDS:

computer vision; artificial intelligence; machine learning; geophysics; neural networks. deep learning.

 

PAGES: 45
BIG AREA: Engenharias
AREA: Engenharia Elétrica
SUMMARY:

Accurately delimiting the transition between continental and oceanic crusts (COB/COT) represents one of the most complex challenges in marine geophysics and has become a topic of growing interest within the scientific community, due to the structural complexity of continental margins and the gradual nature of this transition. Proper mapping of this interface provides essential information on tectonic evolution, sedimentary basin formation, and the architecture of the Earth’s crust, in addition to having relevant scientific, economic, and geopolitical implications, particularly in the context of natural resource exploration and the definition of maritime boundaries. Traditionally, COB/COT delimitation requires a multidisciplinary approach based on the integration of geological, geophysical, and geospatial data, and is strongly dependent on manual interpretation by domain experts. This subjective character highlights the need for more objective, reproducible, and efficient methodologies. In this context, machine learning techniques emerge as a promising alternative for identifying complex patterns and subtle relationships in the data that indicate the presence of the continent–ocean transition. This dissertation proposes the development of an automated methodology for COB/COT mapping using seismic data and convolutional neural networks (CNNs). The approach is based on the classification of seismic image blocks (patches) previously labeled by specialists, considering five geological classes: water, sedimentary basin, continental crust, oceanic crust, and mantle. The dataset was constructed from interpreted seismic images, with the application of data augmentation techniques to balance class distributions. A sequential CNN architecture was designed, trained, and validated, enabling the subsequent spatial reconstruction of the segmented seismic image. The results demonstrate that the model achieved satisfactory performance, with an
overall accuracy of approximately 82% on the test set, showing good discrimination among the main geological units. Higher performance was observed for the identification of water, sedimentary basin, and oceanic crust classes, while the distinction between continental crust and mantle remains challenging, reflecting the intrinsic geophysical similarity between these units. The reconstructed segmented images exhibited spatial coherence
and good agreement with the reference interpretations. Despite existing limitations, the results confirm the feasibility of using deep learning techniques as a support tool for automated COB/COT delimitation. The proposed methodology contributes to reducing interpretative subjectivity and represents an advance in the
application of artificial intelligence to seismic analysis, paving the way for future studies involving larger datasets and diverse geological settings.

 

COMMITTEE MEMBERS:
Presidente - 347628 - ADRIAO DUARTE DORIA NETO
Interno - 2579664 - ALLAN DE MEDEIROS MARTINS
Externo ao Programa - 1315614 - DAVID LOPES DE CASTRO - UFRNExterna à Instituição - ALANNY CHRISTINY COSTA DE MELO