Efficient Channel Attention Based Hybrid Networks for EEG Emotion Recognition

Electroencephalogram (EEG) emotion recognition which utilizes EEG signals to automatically identify and evaluate emotional states in humans, has seen applications of deep learning methods in recent times. In order to effectively utilize the channel features and time series properties of EEG, the efficient channel attention (ECA) based hybrid networks for EEG emotion recognition is introduced in this article. The hybrid networks consist of three parts: the one-dimensional convolutional network (1DCNN) is responsible for extracting features from the EEG data; the residual network embeds the ECA mechanism, which is designed to strengthen the extraction of channel features of the input data; the long short term memory network (LSTM) is employed to capture temporal relationships in the EEG sequences. The adaptive momentum estimation (Adam) optimization algorithm is used to update the parameters of the network, which can automatically adjust the learning rate of each parameter based on previous gradients. In the experiments on the DEAP dataset, ECA based hybrid networks achieves 99.50% and 99.42% accuracy for the binary and four-class tasks, respectively. Compared with previous methods, the proposed method is proved to have better multi classification performance.