CASPIAN JOURNAL
MANAGEMENT AND HIGH TECHNOLOGIES
Hand motions detection in EEG signals
 |
Read | Popov E.Yu., Fomenkov S.A. Hand motions detection in EEG signals // Caspian journal : management and high technologies. — 2015. — №4. — pp. 131-140. |
Popov E.Yu. - post-graduate student, Volgograd State Technical University, 28 Lenin Ave., Volgograd, 400005, Russian Federation, popov.e@hotmail.com
Fomenkov S.A. - D.Sc. (Engineering), Professor, Volgograd State Technical University, 28 Lenin Ave., Volgograd, 400005, Russian Federation, saf@vstu.ru
This paper describes hand motions detection method in 32-component EEG signals. This method is based on using convolution neural network as multi-class classifier. In this paper we proposed and empirically evaluated several architectures of convolutional neural network as well as two versions of activation function for hidden layers neurons, shown advantages of using convolutional neural network for investigating problem. The results suggest that this type of classifier can effectively distinguish characteristic features in the initial EEG signals and provide right values of neural network outputs. Using rectifier linear activation function for hidden layers neurons increases classification quality. Convolutional neural network’s architecture is agile and can be easily modified in width (adding new feature maps on convolutional layers) and in depth (adding new convolutional and pooling layers), thus allowing further improvement of classification quality.
Key words: ЭЭГ, нейрокомпьютерный интерфейс, сверточная нейронная сеть, глубокое обучение, функция активации, классификация, ADADELTA, reclified linear, softmax, перекрестная энтропия, вызванные потенциалы, EEG, brain-computer interface, convolutional neural network