DFrFT-ES model for emotion recognition based on fractional Fourier transform of EEG signals

Emotion recognition with EEG signals is one of hot topics in the fields of human-computer interaction and affective computing. Due to the inherent nonlinear and nonstationary nature of EEG signals, most existing emotion recognition models struggle to effectively capture their intricate time-frequency features, however, it results in daunting challenges for feature extraction and affects recognition accuracy. To address the issue of emotion recognition, this paper proposes a novel DFrft-ES model based on fractional Fourier transform. In the model, the EEG signals are preprocessed by decomposing them into five frequency bands. Moreover, by use of the DFrft-EEG method, fractional Fourier transforms of different orders are applied to each frequency band to obtain fractional domain signals of various orders. With the fractional domain signals, the features such as PSD, DE, DASM, RASM and DCAU are extracted to form the emotion feature vectors. To validate the extracted features on the multimodal emotion database DEAP dataset for emotion recognition, an SVM classifier is designed and trained. Subsequently, the optimal order and best features are selected based on the trained results. The experimental results show that using the DE features extracted with the DFrft-EEG method at the 0.2 order yields the highest classification accuracy of 95.63% for the four emotional regions HVHA, LVHA, LVLA, and HVLA on the arousal-valence plane. This demonstrates that the proposed method has good robustness in performing emotion classification tasks and can effectively improve emotion recognition accuracy.