Aportaciones al diseño de interfaces persona-máquina utilizando señales EEG
- Manuel Ramón Mazo Quintas Director
- Sira Elena Palazuelos Cagigas Co-director
Universidade de defensa: Universidad de Alcalá
Fecha de defensa: 03 de xullo de 2017
- José Luis Pons Rovira Presidente/a
- Luciano Boquete Vázquez Secretario
- Ricardo Ron Angevin Vogal
Tipo: Tese
Resumo
The great advantages of computer systems and intelligent devices, whose presence has increased drastically in recent years, are inaccessible to users who can not interact with the conventional interfaces of these systems. This work is part of the research carried out on alternative interfaces aimed at facilitating the access of all possible users to these technologies. This Ph.D. thesis addresses the design of a brain-computer interface, a human-machine interface based on the acquisition and interpretation of electroencephalographic signals, which aims to establish a direct communication channel between the brain and the computer. Using EEG records, generated voluntarily by a user while performing two mental tasks related to the imagination of the movement of his hands, collected only on two electrodes located on the surface of the scalp, an architecture for recognizing those tasks and translating them into computer actions is proposed. In this work, all the stages of an interface of this type are studied and contributions are made in each of its phases. In the initial preprocessing stage, a fusion of the Fourier and the wavelet transforms is proposed. In the phase of feature extraction and selection, several alternatives are presented, based on the evolution of the algorithms of principal component analysis and partial least squares. Four classification architectures, adapted to the methods of feature selection and extraction, have been detailed in the classification or final translation stage. They are based on calculation of distances between original signals and those recovered by the robust principal component analysis algorithm, on the robust discriminant analysis of the signals transformed by this same technique, on a linear regression algorithm on the latent components obtained by the partial least squares algorithm and, finally, on support vector machines. In order to assess in depth all the proposed contributions, a database of electroencephalographic signals has been built with twelve users’records. A thorough statistical study on all the alternatives designed and their possible parameterizations has been carried out, performing more than 800000 experiments.