• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Doctoral Thesis
DOI
https://doi.org/10.11606/T.55.2020.tde-18032020-095758
Document
Author
Full name
Adam Henrique Moreira Pinto
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Carlos, 2019
Supervisor
Committee
Romero, Roseli Aparecida Francelin (President)
Calvo, Rodrigo
Isotani, Seiji
Ramos, Josue Junior Guimarães
Title in Portuguese
Detecção e análise de sinais EEG com aplicação em robótica educacional
Keywords in Portuguese
Interação homem-máquina
Interfaces cérebro-computador
Robótica educacional
Abstract in Portuguese
Com a tecnologia, existem muitas formas de se aprimorar o aprendizado, mesmo fora da sala de aula. Sistemas educacionais têm sido bastante empregados para essa finalidade, inclusive com o uso de robôs, mas ainda pecam em alguns aspectos de interação com os humanos. As interfaces cérebro-computador (BCI) são sistemas que permitem a comunicação entre usuário e computador a partir de informações do cérebro, podendo dar mais robustez aos sistemas robóticos educacionais. As dificuldades dos alunos são claras durantes provas e outras atividades de avaliação, o problema são os erros durante os estudos para essas provas. Para ajudar neste ponto do aprendizado, foi utilizado um sinal evocado no cérebro relacionado à percepção do erro por um usuário, chamado de Error Related Potential (ErrP), que pode ser medido no EEG, uma forma não-invasiva de BCI. Porém, esses sistemas ainda pecam na qualidade do sinal obtido e na acurácia em encontrar esses momentos de erro. Neste trabalho, foi proposto um sistema de detecção do ErrP, passando pela filtragem, extração de características e classificação do sinal. O pré-processamento do sinal passou por filtros FIR e ICA para limpeza de ruídos e artefatos, foram criados vetores de características com as transformadas de Fourier e as famílias Haar e Daucechies de transformadas wavelets. Para classificação, foram comparadas redes neurais (MLP) e de aprendizado profundo (CNN). Os resultados demonstraram uma acurácia de 96% quando o sinal foi aplicado na base criada, e de 77,23% quando aplicada a toda a rede, mostrando ser promissora para utilização em sistemas educacionais. Além disso, mostrou que a diferença entre as famílias wavelets apresentadas neste trabalho foram pequenas, e que sua escolha pode ser feita considerando o tempo para processamento do sinal. Este trabalho serve como um módulo para um sistema educacional maior, que visa preencher algumas lacunas encontradas nos trabalhos disponíveis.
Title in English
EEG signal detection and analysis with application in educational robotics
Keywords in English
Brain-computer interfaces (BCI)
Human-robot interaction (HRI)
Pedagogical roboticsRI)
Abstract in English
With technology, there are many ways to improve learning, even outside the classroom. Edu- cational systems have long been employed for this purpose, including the use of robots, but there are still a lack in some aspects of human interaction. Brain-computer interfaces (BCI) are systems that allow communication between user and computer from brain information, and can give more robustness to educational robotic systems. Students difficulties are clear during tests and other assessment activities, the problem is errors during the studies and preparation for these tests. To help with learning, a brain-evoked signal related to a users perception of error, called Error Related Potential (ErrP), was used, which can be measured in EEG, a noninvasive form of BCI. However, these systems still lack the quality of the signal obtained and the accuracy of finding these Errp signals. In this work, a ErrP detection system was proposed, including filtering, feature extraction and signal classification. The preprocessing of the signal went through FIR and ICA filters for noise and artifact cleaning, feature vectors were created with the Fourier transforms and the Haar and Daubechies families of wavelet transforms. For classification, neural networks (MLP) and deep learning networks (CNN) were compared. The results showed an accuracy of 96% when the signal was applied to the base created, and 77,23% when applied to the whole database, showing to be promising for use in educational systems. Moreover, it showed that the difference between the wavelet families presented in this work were small, and that their choice can be made considering the time for signal processing. This proposal works as a module for a larger education system, which aims to fill in some of the gaps found in the available work.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Publishing Date
2020-03-20
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.