An emotion recognition method based on EWT-3D-CNN-BiLSTM-GRU-AT model.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Emotion classification from electroencephalogram (EEG) data is crucial for brain-machine interaction (BMI).
• Robustness in emotion recognition systems is a key challenge, addressed by manual feature engineering or AI-driven feature inference.
• EEG signals are inherently noisy, non-stationary, and non-linear, requiring advanced processing techniques.

Purpose of the Study:

• To propose a novel artificial intelligence-based method for enhancing the robustness and accuracy of EEG emotion classification.
• To integrate signal decomposition, feature extraction, and deep learning for a comprehensive emotion recognition framework.
• To combine spatial and temporal information from multichannel EEG recordings for improved classification.

Main Methods:

• Empirical Wavelet Transform (EWT) was used for EEG signal decomposition to obtain frequency components.
• Frequency-based, linear, and non-linear features were extracted and mapped to 2-D and subsequently 3-D representations.
• A 3-D deep learning model incorporating Convolutional Neural Network (CNN), Bidirectional Long-Short Term Memory (BiLSTM), Gated Recurrent Unit (GRU), and self-attention (AT) was developed (EWT-3D-CNN-BiLSTM-GRU-AT).

Main Results:

• The developed EWT-3D-CNN-BiLSTM-GRU-AT model achieved high classification accuracies of 90.57% for valence and 90.59% for arousal on the DEAP dataset.
• The framework successfully combined handcrafted features with state-of-the-art deep learning models.
• Evaluated using a person-independent approach, the model demonstrated superior performance compared to existing emotion classification methods.

Conclusions:

• The proposed novel framework significantly improves the accuracy and robustness of EEG-based emotion classification.
• The integration of EWT, feature engineering, and advanced deep learning architectures offers a powerful approach for affective computing.
• This study provides a strong foundation for developing more sophisticated and reliable emotion recognition systems for BMI and other applications.