Uma metodologia para detecção de sonolência em tempo real com EEG vestível com duas derivações

• Ano de defesa: 2020
• Autor(a) principal: Cassenote, Vanessa
• Orientador(a): Rodrigues, Cesar Ramos
• Banca de defesa: Winck, Ana Trindade, Welfer, Daniel
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Centro de Tecnologia
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação
• Departamento: Ciência da Computação
• País: Brasil
• Palavras-chave em Português: EEG vestível; Processamento de sinais; Aprendizado de máquina
• Palavras-chave em Inglês: Wearable EEG; Signal processing; Machine learning
• Área do conhecimento CNPq: CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.ufsm.br/handle/1/22267

Modern life often requires the subject to modify his circadian cycle due to his profession. This change implies low performance, bad mood and, mainly, reduced attention and, drowsiness at unsuitable times. According to the WHO, about 1.35 million people die each year in traffic accidents. Thus, this work proposes a methodology for detecting drowsiness in real time, with the aim of helping to minimize the problem mentioned. The proposed methodology extracts Alpha, Theta, Beta and Gamma through the Haar Wavelet transform and compares performances of the classifiers: MLP, KNN, LDA, RF, SVM and a Threshold. All analyzes performed used signals from a public database. In order to be able to evaluate the methodology in a more realistic environment, part of the signals were separated and through an experiment, they were acquired by a wearable EEG. From there, the acquired signal and the classifier performances for these signals were analyzed. In addition, the performances of the classifiers for all samples (without EEG aquisition) and different epoch sizes were also evaluated, being a 10 s epoch with a sliding window of 3 s the best, where a sensitivity of 95% was obtained with an SVM.