Detecção e classificação de perturbações no meio aquático via aprendizado profundo e rede neural artificial integrada à métodos estatísticos

• Ano de defesa: 2023
• Autor(a) principal: SOUSA, Monik Silva
• Orientador(a): FONSECA NETO, João Viana da
• Banca de defesa: BARROS FILHO, Allan Kardec Duailibe , FREIRE, Raimundo Carlos Silvério , FONSECA NETO, João Viana da
• 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: análise discriminante linear; detecção de manchas de óleo; machine learning; rede neural artificial; deep learning;
• Palavras-chave em Inglês: linear discriminant analysis; oil stain detection; machine learning; artificial neural network; deep learning.
• Área do conhecimento CNPq: Engenharia Elétrica
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/4645

Due to the growth of oil exploration and transportation, the risk of accidents in the aquatic environment also increases, making it necessary to develop methods and systems that created to reduce the damage caused by industrial activities to the environment. In this dissertation a methodology for detection and classification of disturbances in the aquatic environment is presented. In order to contribute to a solution to the environmental issue and promote scientific and technological advancement through the application of artificial intelligence and statistical methods. Specifically, to detect oil slicks on the surface of the ocean. The developed methodology is based on deep learning approaches, artificial neural network and statistical methods. Based on these approaches, two algorithms were combined for the critic module (performs all exploratory data analysis) of a decision-making system. The first model is a perceptron-type artificial neural network that is integrated with statistical methods, in this case the linear discriminant analysis (LDA) algorithm that defines a discriminant function to estimate the class of images, and the perceptron-type neural network of multiple layers (MLP) to dectact/classify the information, called LDA-MLP model. The second model is just a neural network that uses deep learning, Unet architecture and is called AP-Unet model. To evaluate the performance of ocean oil slick classifieds, information from a Synthetic Aperture Radar (SAR) processed by LDA-MLP and AP-Unet classifieds was used. The database used has 1112 images, 880 of which show oil slicks on the ocean surface, this database is divided into a training set with 1002 images, and a test set with 110 images. With the results obtained and their analysis carried out, it is concluded that the methods of detecting oil slicks on the surface of the ocean, standards, are able to detect oil slicks with good precision, comparing the two methods observing that the two models showed very close accuracy.