Análise de sinais eletroencefalográficos para a classificação de atividades: uma solução via aprendizado de máquina e imagética motora

• Main Author: Nóbrega, Taline dos Santos
• Publication Date: 2020
• Format: Master thesis
• Language: por
• Source: Repositório Institucional da UFRN
• dARK ID: ark:/41046/001300000fkrx
• Download full: https://repositorio.ufrn.br/handle/123456789/32625
• topic: Eletroencefalograma (EEG); Aprendizado de máquina; Random Forest; Imagética motora; Mindwave; V-AMP

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.