High-Resolution Time-Frequency Analysis of EEG Signals for Affective Computing

Affective computing is a critical aspect of human-computer interaction. Electroencephalographic (EEG) signals, which reflect electrical brain activity, are widely used for the understanding of human emotional states. However, these signals are nonlinear and nonstationary, making traditional analysis methods insufficient. To address these challenges, recent studies have focused on time-frequency analysis. In this paper, we propose a variable frequency complex demodulation (VFCDM) approach to obtain high-resolution time-frequency spectra (TFS) from EEG signals. First, we compute the TFS using the time-varying optimal parameter search technique to capture the spectral information. Then we generate VFCDM sub-bands and extract statistical features from each of the sub-bands. These features are then used with the Random Forest algorithm to classify arousal and valence dimensions. Our results demonstrate the robustness of this approach and its ability to accurately discriminate complex affective dimensions. The δ-VFCDM and γ-VFCDM bands produced the highest F1 scores of 71.80% for Arousal and 69.55% for Valence differentiation. This work significantly advances EEG-based affective computing and opens avenues for more emotionally attuned human-computer interaction systems.