EEG-based emotion recognition using deep learning techniques

Abstract

With increasing development and growth of BCI (brain-computer Interaction) technology, the emotion recognition technology based on EEG (Electroencephalograph) grew mature in recent years. In this dissertation report, a literature review of EEG-based BCI system is presented. The basic structure of a EEG-based emotion recognition system is illustrated. Several significant experiments which promoted the research and a number of important algorithms constitute the emotion recognition system are recorded. Further more, a novel emotion identification framework on the strength of EEG signals is put forward. 32 channels DEAP database is applied and processed with 1s Hanning window. PSD (Power Spectrum density) and PCC (Pearson’s Correlation Coefficients) are chosen to be the extracted features. PCA algorithm is used to reduce the demensionality of the feature sequence. In baseline experiment, the features are fed to SVM classifier and the average recognition accuracy is 58.09% in valence and 63.27% in arousal. Meanwhile, In proposed experiment, standard LSTM neural network is applied and this experiment gets 70.44% in valence and 67.36% in arousal as average recognition accuracy. Whats more, some tests aimed at optimization of the EEG-based emotion recognition are conducted. The deeper LSTM neural network structure and BiLSTM neural network are applied in subsequent studies.