A novel automated oil spill detection approach based on the q-Exponential distribution and machine learning models

Detalhes bibliográficos
Ano de defesa: 2022
Autor(a) principal: NEGREIROS, Ana Cláudia Souza Vidal de
Orientador(a): LINS, Isis Didier
Banca de defesa: Não Informado pela instituição
Tipo de documento: Tese
Tipo de acesso: Acesso aberto
Idioma: eng
Instituição de defesa: Universidade Federal de Pernambuco
Programa de Pós-Graduação: Programa de Pos Graduacao em Engenharia de Producao
Departamento: Não Informado pela instituição
País: Brasil
Palavras-chave em Português:
Link de acesso: https://repositorio.ufpe.br/handle/123456789/46587
Resumo: Oil spills are among the most undesirable events in coastal environments because they are substantially harmful, with negative environmental, social, and economic consequences. In general, a risk framework for the event involves prevention, monitoring, detection, and damage mitigation. Regarding detection, rapid oil spill identification is essential for problem mitigation, which generally fosters the use of automated procedures. Usually, automated oil spill detection involves radar images, computer vision, and machine learning techniques to classify these images. In this work, we propose a novel image feature extraction method based on the q-Exponential probability distribution, named q-EFE. Such a probabilistic model is suitable to account for atypical extreme values of the variable of interest, e.g., pixels values, as it can have the power-law behavior. The q-EFE part is combined with machine learning methods to comprise a computer vision methodology to automatically classify images as “with oil spill” or “without oil spill”. Hence, we also propose a new automatic oil spill detection methodology that uses the q-EFE to rapidly identify oil spills in radar images. We used a public dataset composed of 1112 Synthetic Aperture Radar (SAR) images to validate our proposed methodology. Considering the proposed q-Exponential-based feature extraction, the tested Machine Learning methods and Deep Learning models architectures, the Support Vector Machine (SVM) and Extreme Gradient Boosting (XGB) models outperformed deep learning models and Local Binary Pattern (LBP) and Gray Level Co-occurrence Matrix (GLCM) techniques for the biggest dataset size.