ABSTRACT
Electroencephalograms (EEG) are the complex signals captured using the electrodes attached to human scalp. Motor imagery (MI) EEG signals are the class of EEG signals that helps in intents recognition for accurate working of brain computer interface (BCI) systems. These signals provide information about the bioelectrical signal fluctuations that BCI systems use to predict what the human brain is thinking or imagining. The latest technologies like machine learning and deep learning frameworks are used in identification and classification of the MI-EEG signals. In this chapter, fusion of two deep learning frameworks is proposed to efficiently extract the MI-EEG signal features, which are then used for classification using various machine learning algorithms. The deep feature learning framework consists of two major network structures, namely recurrent neural network (RNN) and convolutional neural network (CNN). RNN is used to extract a temporal feature vector, and CNN is used extract a spatial feature vector from the input MI-EEG signals. The two different feature vectors are then stacked into single feature vector and are used to train different machine learning classification algorithms. Out of many algorithms tested, the support vector machine (SVM) achieved high performance in less time. The publicly available Physionet MI-EEG signals are used to evaluate the proposed methodology. This end-to-end architecture can achieve MI-EEG multiclass classification accuracy of about 98.16% with very high convergence speed as compared to existing state-of-the-art methods.
