SSVEP-EEG signal pattern recognition system for real-time brain-computer interfaces applications

• Autor(a) principal: Giovanini, Renato de Macedo [UNESP]
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/151710
• topic: Pattern recognition; Machine learning; Brain-machine interface; Brain-computer interface; Python; Raspberry Pi; Open-source; Reconhecimento de padrões; Aprendizado de máquina; Interface cérebro-máquina; Interface cérebro-computador

There are, nowadays, about 110 million people in the world who live with some type of severe motor disability. Specifically in Brazil, about 2.2% of the population are estimated to live with a condition of difficult locomotion. Aiming to help these people, a vast variety of devices, techniques and services are currently being developed. Among those, one of the most complex and challenging techniques is the study and development of Brain-Computer Interfaces (BCIs). BCIs are systems that allow the user to communicate with the external world controlling devices without the use of muscles or peripheral nerves, using only his decoded brain activity. To achieve this, there is a need to develop robust pattern recognition systems, that must be able to detect the user’s intention through electroencephalography (EEG) signals and activate the corresponding output with reliable accuracy and within the shortest possible processing time. In this work, different EEG signal processing techniques were studied, and it is presented the development of a EEG under visual stimulation (Steady-State Visual Evoked Potentials - SSVEP) pattern recognition system. Using only Open Source tools and Python programming language, modules to manage datasets, reduce noise, extract features and perform classification of EEG signals were developed, and a comparative study of different techniques was performed, using filter banks and Discrete Wavelet Transforms (DWT) as feature extraction approaches, and the classifiers K-Nearest Neighbors, Multilayer Perceptron and Random Forests. Using DWT approach with Random Forest and Multilayer Perceptron classifiers, high accuracy rates up to 92 % were achieved in deeper decomposition levels. Then, the small-size microcomputer Raspberry Pi was used to perform time processing evaluation, obtaining short processing times for every classifiers. This work is a preliminary study of BCIs at the Laboratório de Instrumentação e Engenharia Biomédica, and, in the future, the system here presented may be part of a complete SSVEP-BCI system.