Annotation

Annotation: The article discusses some methods of analyzing the electroencephalographic signal (EEG signal) in order to identify the characteristic features in it for their subsequent classification. The possibility of classifying movement EEG data by two subjects of different groups of facial muscles is investigated. To receive electroencephalograms and transmit them for processing, Emotiv EPOC+ equipment with a wireless interface for connecting to a computer and the possibility of 14-channel recording was used. The d ata analysis was based on the results of 150 sessions of 16 seconds each, in which 5 different reactions were considered. The paper considers and compares various methods of extracting features from an EEG signal, namely: the coefficients of the Fourier sp ectral decomposition, the values of the Alter-Johnson function, SAX-BOP features, linear correlations, and classical statistics. Testing is carried out in the framework of two neural network models: convolutional and fully connected networks. In addition, methods based on direct analysis of statistical information are also considered. According to the results of the study, estimates of the accuracy of detecting the tested reactions with various methods of extracting features were obtained. The best results were found for methods based on the spectral Fourier decomposition on a fully connected network-about 95% and 90% on the training and test samples, and using the values of the Alter-Johnson function on the convolutional network-about 99% and 86%.