masterThesis
Análise de sinais eletroencefalográficos para a classificação de atividades: uma solução via aprendizado de máquina e imagética motora
Fecha
2020-01-31Registro en:
NÓBREGA, Taline dos Santos. Análise de sinais eletroencefalográficos para a classificação de atividades: uma solução via aprendizado de máquina e imagética motora. 2020. 68f. Dissertação (Mestrado em Engenharia Elétrica e de Computação) - Centro de Tecnologia, Universidade Federal do Rio Grande do Norte, Natal, 2020.
Autor
Nóbrega, Taline dos Santos
Resumen
The human body motor activities, as well as those activities related to decision
making, emotional, and psychic issues, can be understood by analyzing the electrical
signals from the brain, also known as electroencephalogram (EEG) signals. The study
and application of these data have been growing within the scientific community. The
use of these signals has contributed to the development of Brain Computer Interfaces
(BCI), which represents the future of assistive technologies, especially for people who
do not have motor control. However, the extraction of characteristics and patterns of
these signals is still a complicated process. Machine learning algorithms have been
showing excellent results for EEG signals interpretation, and they are also useful as a tool
for classification and analysis. Their applications involve neuroscience studies, neural
engineering, and even commercial uses. Thus, the purpose of this paper is to analyze the
signals from the neural activity of individuals submitted to protocols involving motor and
imagery tasks, in order to propose a classifier for such tasks. Imaging tasks, specifically
motor imagery, can be understood as neurocognitive techniques that the subject imagines
performing a motor action without performing the proper movement. For example, it is
a mental process in which the person imagines the movement of the body but do not do
it. The interpretation and classification of this type of signal allow the development of
control tools that can be activated through cognitive processes. The sensors used were
a 16-channel electroencephalogram and a low-cost one-electrode sensor with wireless
connection technology. The proposed classification solution is based on Random Forest
machine learning technique. For both sensors, the proposed algorithm proved to be
efficient in the process of identifying the type of movement (real or imaginary) and what
limb performed it (hands or ankles right and left).Additionally, it was also possible to
validate some difficulties already pointed out by other researchers in the area, such as the
expressive interpersonal variability of EEG signals, which contributes negatively to the
classification process.