Emotion Recognition based BCI using Channel-wise Features

Abstract

Electroencephalogram (EEG) based emotion recognition has received significant research attention since it allows the direct reflection of the inner state of the brain activity. However, it is difficult to improve the performance of recognizing cross-subject emotion due to subjectivity and the noise associated with EEG data collection. In this paper, we propose a novel cross-subject EEG-based emotion recognition method with combination of channel-wise features and LSTM. The channel-wise feature is defined by the symmetric matrix, the element of which is calculated by the Pearson correlation coefficient between two-pair channels to consider the spatial interaction among all channels. Then, the channel-wise features are fed to the 2-layer stacked Long Short-Term Memory (LSTM), which can extract temporal features and learns an emotional model that can complementarily handle subjectivity and noise. The experiments on two publicly available datasets DEAP and SEED demonstrate the effectiveness of the combined use of channel-wise features and LSTM. Experimental results achieve state-of-the-art classification rate of 98.93% and 99.10% over 2-classes valence and arousal on DEAP respectively and show the accuracy of 99.63% over 3-classes classification on SEED.