A Hybrid EEG-based Emotion Recognition Approach Using Wavelet Convolutional Neural Networks and Support Vector...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Deep learning and Convolutional Neural Networks (CNNs) are increasingly used in biomedical engineering.
• Integrating CNNs with machine learning methods can enhance accuracy in complex processing tasks.

Purpose of the Study:

• To propose a hybrid approach for improved emotion recognition from electroencephalogram (EEG) signals.
• To leverage deep features from wavelet CNNs (WCNNs) and multiclass support vector machines (MSVM) for enhanced classification.

Main Methods:

• EEG signals were preprocessed and converted to time-frequency scalograms using continuous wavelet transform (CWT).
• Pre-trained CNNs (AlexNet, ResNet-18, VGG-19, Inception-v3) were fine-tuned, and deep features were extracted.
• Extracted features were classified using MSVM, with evaluation on DEAP and MAHNOB-HCI databases using a subject-independent leave-one-subject-out criterion.

Main Results:

• Extracting deep features from ResNet-18's earlier convolutional layer (Res2a) and using MSVM classification improved average accuracy, precision, and recall by approximately 20% and 12% on MAHNOB-HCI and DEAP databases.
• Combining scalograms from specific brain regions (pre-frontal, frontal, parietal, parietal-occipital) achieved higher average accuracies of 77.47% (MAHNOB-HCI) and 87.45% (DEAP).

Conclusions:

• The combination of CNN and MSVM significantly enhanced emotion recognition from EEG signals.
• The proposed hybrid method achieved results comparable to state-of-the-art studies in the field.