Abstract In recent years, deep learning-based emotion recognition from electroencephalography (EEG) signals has garnered significant attention in brain-computer interfaces. However, effectively capturing local and global dependencies remains a challenge due to the complexities of EEG data. Furthermore, traditional convolutional neural networks and RNNs often struggle to fully explore the spatio-temporal relationships between different features. To address these issues, we propose an end-to-end model with the augmented capsule-gated Transformer to improve the performance of EEG emotion recognition, in which we learn cross-channel spatial features effectively, and the raw EEG signals are automatically weighted to emphasize key attributes. Subsequently, the capsule network extracts low-level and high-level spatial information, fully leveraging the potential insights within the signals. Building on this, an efficient Transformer is employed to model the relationships among different electrodes, allowing for a more in-depth analysis of the temporal dependencies across multiple features. Extensive experiments are conducted on the Dataset for Emotion Analysis using Physiological Signals (DEAP) dataset, and comparison results with existing state-of-the-art methods demonstrate the superior performance of the proposed method. Specifically, for the arousal and valence dimensions, the average recognition accuracies in subject-dependent experiments reach 93.51% and 94.24%, while the subject-independent experiments achieve average accuracies of 86.78% and 87.59%.
Wang et al. (Sat,) studied this question.