A Deep Learning Method for Classification of EEG Data Based on Motor Imagery

Abstract

Effectively extracting EEG data features is the key point in Brain Computer Interface technology. In this paper, aiming at classifying EEG data based on Motor Imagery task, Deep Learning (DL) algorithm was applied. For the classification of left and right hand motor imagery, firstly, based on certain single channel, a weak classifier was trained by deep belief net (DBN); then borrow the idea of Ada-boost algorithm to combine the trained weak classifiers as a more powerful one. During the process of constructing DBN structure, many RBMs (Restrict Boltzmann Machine) are stacked on top of each other by setting the hidden layer of the bottom layer RBM as the visible layer of the next RBM, and Contrastive Divergence (CD) algorithm was also exploited to train multilayered DBN effectively. The performance of the proposed DBN was tested with different combinations of hidden units and hidden layers on multiple subjects, the experimental results showed that the proposed method performs better with 8 hidden layers. The recognition accuracy results were compared with Support vector machine (SVM) and DBN classifier demonstrated better performance in all tested cases. There was an improvement of 4 – 6% for certain cases.