Adaptive construction of critical brain functional networks for EEG-based emotion recognition

In electroencephalography (EEG) emotion recognition research, acquiring excellent discriminative features is essential for improving classification performance. Features of brain functional networks carry critical multidimensional information and provide a comprehensive understanding of the coordination among brain regions. Existing emotion recognition models based on brain networks are overly complex and affected by individual emotional differences. Such complexity and variability hinder the practical application of real-time emotion recognition systems. This paper proposes a lightweight construction strategy for brain networks to recognize emotions. The adaptive emotion sample window selection (AESW) method identifies windows with the most informative emotional segments, improving EEG data processing and reducing signal length. The stability brain connectivity determination (SCD) method dynamically ascertains the key connectivity features for each individual by analyzing changes in the brain network across different emotional states. The emotional traits constructed for each subject’s adaptation preserve the diversity of temporal correlations among different brain regions. In addition to validation on public datasets, extensive testing was conducted on the proposed large-scale emotion dataset, MuSer. The results demonstrated a notable improvement in emotion recognition performance compared to traditional methods. This method holds the potential for real-time construction of brain functional networks.