Fusion Network Modeling for Cross-Time Emotion Recognition from EEG

In recent years, emotion recognition technology has played an important role in various industries. Most of the publicly available cross-temporal EEG model focus on extracting temporal or spatial features for emotion classification, while neglecting the integration of the brain's overall structure with multidimensional information. Thus, methods based on manual feature extraction perform poorly in cross-temporal EEG emotion recognition. To address these issues, this paper proposes a novel cross-temporal EEG emotion recognition model. The model consists of a graph convolutional network and a convolutional neural network with an attention mechanism. The graph convolutional network naturally fits the brain's structure and provides both temporal and spatial information based on channel positions. The convolutional neural network with attention mechanism explores important and discriminative features on top of enhancing the representational capacity of deep networks. Five experiments were conducted on 24 subjects, with time intervals of 1 day, 3 days, 7 days, and 14 days, resulting in a dataset of 720 cases of EEG data encompassing three emotions. The data from the initial three experiments constituted the training set, while the data from the final two experiments served as the test set. By integrating multiple networks into a unified architecture, we demonstrate enhanced classification performance. Our method achieved an average classification accuracy of 62.79%.in the datasets used in this study. The experimental results indicate that the fused model can comprehensively and flexibly learn and process EEG signal data, providing deep and shallow features required for classification.