An Improved Dual Stream DGCNN Method for Fine-Grained Positive Emotion Classification

Recent advances in EEG-based emotion computation have garnered significant interest from the fields of neuroscience and computer science. Despite this, the bulk of EEG-based emotion classification research has concentrated on the dimensions of arousal and valence, with scant attention to the classification of positive emotions as experienced in daily life. Addressing this gap, our study focuses on the classification of four specific positive emotions using the open-source FACED dataset. We processed EEG data by segmenting it, extracting features, and then designing and training the DualStreamDGCNN deep learning model, which achieved the best accuracy of 60.88%. Furthermore, we visualized the adjacency matrices of the four emotions from our model, offering insights from a neuroscientific perspective. This research is pivotal for advancing our understanding of the types and dynamics of positive emotions in practical applications, such as enhancing educational environments and boosting employee efficiency.