Objective Emotion recognition is an important part for affective computing. With the development of computer science and technology, the application demand for intelligent human-computer interaction and affective brain-computer interface is also increasingly urgent. To expand the applications of human-computer interaction, electroencephalogram-based (EEG) emotion recognition, which aims to enhance the computers with the ability to perceive and generate human emotions, is the hotspot for affective computing. Essentially, for EEG-based emotion recognition, the performance of EEG-based emotion decoding mainly depends on the effectiveness of emotional EEG feature extraction. In this work, we proposed an efficient feature extraction strategy based on emotional EEG brain networks to further improve the robust performance of EEG-based emotion recognition.

Methods In the proposed framework, original emotion EEG signals were first used to construct corresponding emotional brain networks with Phase Locking Value (PLV). And then, the constructed emotional EEG brain networks were represented in a low-dimensional subspace searched from the original brain network vector space, and a set of brain network features were extracted from the corresponding brain networks. With the extracted brain network features, the emotion recognition systems were trained with the “Libsvm toolbox” in MATLAB 2018b to realize the emotion recognition with the brain network sample from testing emotional EEGs.

Results To evaluate the performance of the proposed model, comparison experiments were conducted on three public emotional EEG datasets (SEED, DEAP, MAHNOB-HCI) with the proposed emotion recognition model and other existing methods. The experimental results have shown that the proposed model has achieved a more robust performance. Specifically, the proposed model has achieved the results of 95.01% and 85.17% on the SEED and MAHNOB-HCI datasets with the three-emotion classification tasks (Natural, Negative and Positive), respectively. Besides, for the four-emotion classification task (HAHV, HALV, LAHV, and LALV), the proposed model has achieved 83.86% on DEAP dataset. Compared with other existing methods, the proposed model has consistently achieved robust performance and the superiority of the proposed model has been verified.

Conclusion Compared with other existing methods, in this work, emotional EEG brain networks were adopted for emotion recognition and robust results have been achieved on three public datasets in the off-line experimental conditions. The proposed emotional brain network-based emotion recognition model may provide a potential solution for the realization of intelligent human-computer interaction and affective brain-computer interface.