Table 5 Emotion recognition performance of different models
From: EEG emotion recognition based on differential entropy feature matrix through 2D-CNN-LSTM network
Reference index | Feature extraction methods | Classifier used | Database | Accuracy(%) |
|---|---|---|---|---|
Yoon et al. [37] | FFT-based spectral power features extracted from EEG rhythms | Bayesian | DEAP | Arousal: 0.709 Valence: 0.701 |
Arnau-González et al. [38] | Spectral power, energy, and connectivity features | SVM | DREAMER | Arousal: 0.862 Valence: 0.854 |
Gupta et al. [39] | Information potential feature extracted in the FAWT domain of EEG signal | Random forest | DEAP | Arousal: 0.714 Valence: 0.799 |
Gupta et al. [40] | Graph–theoretic-based EEG features | RVM | DEAP | Arousal: 0.67 Valence: 0.69 |
Cheng et al. [41] | 3D cube evaluated from EEG segment | CNN | DEAP | Arousal: 0.894 Valence: 0.904 |
Soleymani et al. [42] | EEG power | SVM | MAHNOB-HCI | Arousal: 0.52 Valence: 0.57 |
S. Katsigiannis, et al. [43] | Power spectral density-based features extracted from EEG signal | SVM | DREAMER | Arousal: 0.624 Valence: 0.621 |
Zhang et al. [44] | Temporal slices obtained from each channel EEG signal | Recurrent attention model | DREAMER/DDEAP | Arousal: 0.855 Valence: 0.836 |
Yin et al. [45] | EEG’s differential entropy | ERDL | DEAP | Arousal: 0.848 Valence: 0.852 |
Topic et al. [46] | TOPO-FM | CNN + SVM | DEAP | Arousal: 0.806 Valence: 0.857 |
Liu et al. [47] | DE, statistical features | DCAA | DEAP | Arousal: 0.843 Valence: 0.856 |
Yang et al.[48] | Row signals | PCRNN | DEAP | Arousal: 0.913 Valence: 0.908 |
Gao et al.[49] | Time domain and frequency domain | SVM | DEAP | Arousal: 0.752 Valence: 0.805 |
Our method | EEG’s differential entropy | 2D-CNN-LSTM | DEAP | Arousal: 0.919 Valence: 0.923 |