An EEG emotion recognition method based on shallow feedback spiking neural network

With the application of brain-computer interaction technology in various fields, emotion recognition based on EEG has been widely studied. However, the existing artificial design of emotional feature extraction methods are not universal, and the deep model is not highly interpretable and easily falls into a bottleneck, resulting in the inability to carry out end-to-end emotion recognition on embedded devices. To solve this problem, a spiking neural network with shallow feedback structure is proposed and applied to emotion EEG recognition and classification. It is built with adaptive threshold LIF (Leaky Integrate and Fire) neurons combined with reward STDP (spike-timing-dependent Plasticity) learning rules, and the spiking neural network simulates the information encoding and processing mechanism in the human brain. The spatial and temporal characteristics of the data can be extracted efficiently, and the network connection between the processing layers in the network can form a side suppression mechanism, which can improve the network adaptive ability. The proposed method finally achieved 92.6% and 90.3% accuracy on DEAP and SEED-IV, respectively, which increased the accuracy by 8.1% and 5.4% compared with the deep learning residual network with the same setting. The model was also greatly compressed, thus providing a reference solution for the real-time application of brain-computer interaction systems.