Determining Levels of Affective States with Riemannian Geometry Applied to EEG Signals

Emotion recognition from electroencephalography (EEG) has drawn intense interest, yet most work still relies on Euclidean feature spaces that ignore the curved geometry of covariance matrices. We introduce the a pipeline that comprises Riemannian manifold that, when combined with the Fisher Geodesic Minimum-Distance-to-Mean (FgMDM) classifier, leverages the full Riemannian structure of symmetric positive-definite (SPD) EEG covariances. Our approach applies an additional geodesic-mean filter that fuses information across channels and trials. Experiments on the five-class SEED-V dataset show that the proposed pipeline achieves high classification accuracy and demonstrates improved robustness and interpretability compared to other approaches and baselines. The results confirm the potential of Riemannian geometry as a powerful framework for emotion recognition tasks, especially when dealing with high-dimensional and non-stationary EEG data.