Human emotion recognition based on multi-channel EEG signals using LSTM neural network

Electroencephalogram (EEG) signal is often used in emotion recognition tasks to classify human emotions. In this paper, we propose a new approach to learn the temporal features of EEG using long and short-term memory (LSTM), which is a type of Recurrent Neural network (RNN), especially suitable for solving the problem of long-term dependencies such as gradients vanishing and exploding. In addition, to enhance the interaction between EEG signals and to learn the non-linear characteristics between EEG electrodes, we use 1D-Convolution kernel to pre-process the input EEG data. To justify the capability of this method, we set the subject-independent experiments via adopting the leave-one-out experimental strategy on SEED dataset. The result of our experiments shows that this method can effectively capture the timing relationships in EEG signals with high classification accuracy around 93%.