EEG Recognition of Epilepsy Based on Spiking Recurrent Neural Network

Electroencephalogram (EEG) is an important means of epilepsy diagnosis. Deep learning methods can accurately extract EEG information, but the large number of model parameters makes the model difficult to be deployed and applied in edge handheld devices. Therefore, this paper proposes a lightweight epilepsy recognition network. Firstly, the Step-Forward coding method is used to encode the original EEG signal. Then, the replacement gradient method is used to train the Spiking Recurrent Neural Network (SRNN) to analyze the influence of neurons and replacement functions on the model. Finally, the SRNN model built by adaptive neurons is built. The accuracy, sensitivity and specificity of the three classification task are 97.05%, 96.79% and 99.51%, respectively. Compared with other methods, this method achieves better classification results with fewer parameters.