Classificação de sedimentos da margem equatorial brasileira utilizando algoritmos

• Ano de defesa: 2024
• Autor(a) principal: LOPES, Luiz Eugênio Hoffmann
• Orientador(a): BARROS FILHO, Allan Kardec Duailibe
• Banca de defesa: BARROS FILHO, Allan Kardec Duailibe , BARREIROS, Marta de Oliveira , SANTANA, Ewaldo Eder Carvalho , ROCHA, Priscila Lima
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
• Departamento: DEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCET
• País: Brasil
• Palavras-chave em Português: aprendizado de máquina; classificação; margem equatorial brasileira; sedimentos marinhos.;
• Palavras-chave em Inglês: Classification; brazilian equatorial margin; machine learning; marine sediments.
• Área do conhecimento CNPq: Eng
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/5313

Recently, important discussions about the occurrence of extensive carbonate reefs and the potential for hydrocarbon discoveries in the Pará-Maranhão and Foz do Amazonas basins have raised the importance of studying the faciological composition of the continental shelf. However, sampling marine sediments is a high-cost and often time-consuming operation, depending on the distance from the coast and the depth of sampling, requiring the displacement of staff and equipment through specialized research vessels. With this in mind, the main objective of this work was to integrate sedimentological information from bottom samples and seismic attributes from machine learning algorithms to optimize the mapping of facies on the continental shelf of the Brazilian Equatorial Margin. Sediment data were acquired from the National Oceanographic Data Bank (BNDO), classified as Mud, Sand and Gravel for grain size and Lithoclastic (<30%), Lithobioclastic (<50%), Biolithoclastic (<70%) and Bioclastic (>70%) for calcium carbonate content. The seismic attributes used were Coherence, Amplitude, Intensity, Instantaneous Phase and Instantaneous Frequency. Five machine learning algorithms were developed in order to relate sediment types and carbonate content to seismic attributes. The results of each algorithm are: Support Vector Machine: Accuracy 80.31%, Precision 81.21%, Recall 80.31% and F1_Score 80.01%. K-Nearest Neighbors: Accuracy 77.07%, Precision 80.55%, Recall 77.07% and F1_Score 75.94%. Decision Trees: Accuracy 85%, Precision 85.78%, Recall 85% and F1_Score 84.85%. Multilayer Perceptron: Accuracy 82.74%, Precision 82.74%, Recall 82.74% and F1_Score 82.49%. Adaptive Boosting: Accuracy 70.47%, Precision 72.25%, Recall 70.47% and F1_Score 69.75%. The best result was obtained from the Decision Trees algorithm with 85% accuracy. With these preliminary results, it is possible to create a machine learning model to automatically classify marine sediments, enabling and optimizing future research on carbonate facies and mesophotic reefs in the Brazilian Equatorial Margin.