A Method for EEG Contributory Channel Selection Based on Deep Belief Network

In order to obtain better performance in BCI systems, multi-channel electrodes are often used to collect EEG signals. However, using multi-channel electrodes may cause inconvenience to the EEG signal acquisition work, and may cause problems such as slow system operation and poor performance. This paper proposes a new contributory channel selection method based on data driven method, which realizes the optimal selection of channels by means of the Deep Belief Network with strong learning ability for high-dimensional vectors. First, the DBN model is trained through the continuous adjustment of the parameters, which result in an optimal DBN model. Then, the distribution of the weights in the first layer of the obtained optimal DBN model are analyzed and the channels with larger weights are selected as the optimal channel combination to achieve the purpose of channel selection. The experimental results show that there are different channel selection results among individuals, and the EEG classification accuracy similar to or higher than that of using high-density channels can be obtained by using selected fewer channels, which enhances the practicability of the BCI system.