Spiking Neural Network Classification Method in Motor Imagery Brain-Computer Interface

Abstract

Electroencephalogram (EEG) Brain-computer interface (BCI) classification research has become a hot field recently. This paper proposes spiking neural network (SNN) classification method of motor imagery EEG signals. Firstly, the raw EEG signal is preprocessed by band-pass filtering, and encoded into spike signals by Ben's Spike encoding algorithm. Then, convolution layers and pooling layer is designed in SNN to take advantage of the characteristics of EEG signal. Finally, the SNN is employed to classify the spike signals. It is worth noting that the activation function in SNN is different from the convolutional neural network. It simulates biological encode information of brain in the form of spikes by the LIF model. By simulating the change process of membrane potential, it realizes the transmission of intercellular spikes and the active release of membrane potential without input, which makes the signal processing closer to the neuron cell membrane potential processing mechanism. On the other hand, since spike signals can weaken the negative effect of noise in the superposition process, spikes input can reduce the influence of noise in the network. The BCI Competition IV 2a datasets are used in the experiments, and the classification results show that the proposed classification method achieves the average classification accuracy of 82.08%.