Uso de Redes Neurais ARTMAP Nebulosas para a classificação de padrões em sinais ECoG relacionados ao movimento dos dedos

• Ano de defesa: 2015
• Autor(a) principal: Monteiro, Rhycardo Luiz [UNESP]
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Redes neurais (Computação); Processamento de sinais; Reconhecimento de padrões; Neural networks (Computer science)
• Link de acesso: http://hdl.handle.net/11449/134102; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/14-01-2016/000857556.pdf

The pattern recognition signals applied to the brain is essential for the implementation of computational interfaces allowing control devices to aid people with physical limitations. The identification of features associated with body movements of the members, such as the fingers, requires a sequence of steps which includes the acquisition and pre-processing of signals, extraction of features and classification of signal data. These signals, called ECoG can be obtained directly from the brain through implants in the region that generates the motion decisions, which is the primary motor cortex. Such signs are superior in information qualitatively and quantitatively compared to the known EEG signals obtained on the surface of the scalp. The pre-processing consists in preparation of the signals to be processed through the relevant channel selection techniques, windowing and filtering for selecting frequency band information carrier. The feature extraction can be done by using these signals in the frequency domain and then subjecting them to autoregression. The classification is made using artificial neural networks ARTMAP-Fuzzy type, having as input matrices composed of processed data from the ECoG signals and data glove, both obtained from the same subject during the experimental section. This work could ultimately, generate sleeve signals from the ECoG signals. The average correlation coefficient obtained was 0.91, showing the efficiency of the proposed model