Multi-channel EEG-based emotion recognition via a multi-level features guided capsule network.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Deep learning (DL) and convolutional neural networks (CNNs) show promise for electroencephalograph (EEG)-based emotion recognition.
• Existing CNN methods often require complex feature pre-extraction and struggle to capture inter-channel relationships in EEG signals.
• Recognizing emotions from EEG is challenging due to the complex nature of brain signals.

Purpose of the Study:

• To propose an effective multi-level features guided capsule network (MLF-CapsNet) for multi-channel EEG-based emotion recognition.
• To develop an end-to-end framework that simultaneously extracts features and determines emotional states from raw EEG signals.
• To improve upon existing methods by better characterizing the intrinsic relationships among different EEG channels.

Main Methods:

• Developed an end-to-end Multi-Level Features Guided Capsule Network (MLF-CapsNet).
• Incorporated multi-level feature maps from different layers to form primary capsules, enhancing feature representation.
• Utilized a bottleneck layer to reduce parameters and computational speed.

Main Results:

• Achieved high average accuracies: 97.97% (valence), 98.31% (arousal), and 98.32% (dominance) on the DEAP dataset.
• Attained accuracies of 94.59% (valence), 95.26% (arousal), and 95.13% (dominance) on the DREAMER dataset.
• Demonstrated superior performance compared to state-of-the-art methods in EEG emotion recognition.

Conclusions:

• MLF-CapsNet effectively recognizes emotions from multi-channel EEG signals in an end-to-end manner.
• The proposed method overcomes limitations of traditional CNNs by better utilizing inter-channel information.
• The results indicate MLF-CapsNet is a highly accurate and efficient approach for EEG-based emotion recognition.