EEG Channel Selection Based on Neuron Proportion with SNN for Motor Imagery Classification

The brain computer interface (BCI) technology based on motor imagery has great potential for various control and communication tasks. However, the presence of a large number of EEG channels leads to redundant information, which affects processing speed and classification accuracy. Spiking neural networks (SNN) have the potential to process EEG data by transmitting pulsing activity between synapses and neurons situated in space. Neucube is an SNN architecture inspired by the human brain structure that allows for end-to-end learning, classification, and understanding of spatiotemporal data at low power consumption, saving computing power and reducing operational complexity. By utilizing this model, the temporal and spatial information of EEG signals can be considered to explore the importance and correlation of spatial neurons corresponding to EEG channels during the classification process. Thus, this study aimed to use the Neucube model based on SNN to select the most influential EEG signal channels in the classification process. This improvement mainly focuses on improving classification accuracy and reducing energy consumption to enhance the practical application performance of BCI systems. The proposed method was tested on the BCI Competition IV Dataset 2A. After deleting several unimportant EEG channels, the classification accuracy was improved, and the energy consumption was reduced.