A method of EEG emotion recognition based on residual contraction network

The emotional constantly affects human body, mind and behavior. The accurate recognition of emotion has a broad application prospect in the medical field. Although deep learning algorithms have made some achievements in emotion recognition research, there are still some problems such as insufficient extraction of EEG (Electroencephalogram) frequency features, spatial features and time features, noise interference and complex model parameters. In this thesis, the 4D-RSC-BiLSTM (Four-Dimensional Residual Shrinkage Convolutional Bidirectional Long Short Term Memory Neural Network) is constructed, and the convolutional layer is inserted into the deep separable convolutional network to complete the improvement of network structure and parameter adjustment. Then the residual shrinkage module is combined with the improved deep separable convolutional network to remove redundant features and reduce the network computing cost while solving the problem of model gradient disappearance. The experimental results show that the proposed model is feasible in emotion recognition.