Multidimensional Feature in Emotion Recognition Based on Multi-Channel EEG Signals.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Electroencephalogram (EEG) signal analysis is crucial for understanding mental states and advancing artificial intelligence.
• Existing methods often struggle to fully capture the spatial and temporal dynamics inherent in EEG data.
• Accurate emotion recognition from EEG is a significant challenge with broad applications.

Purpose of the Study:

• To propose a novel method for EEG-based emotion recognition that preserves spatial information and mines temporal features.
• To develop an efficient neural network architecture for comprehensive EEG signal analysis.
• To enhance the accuracy of emotion classification using deep learning models.

Main Methods:

• Utilized a multidimensional feature structure to input frequency, spatial, and temporal information from multichannel EEG signals.
• Developed a depthwise separable convolution neural network to extract frequency and spatial features efficiently, reducing computational parameters.
• Employed an ordered neuronal long short-term memory (ON-LSTM) network to learn hierarchical information and extract deep emotional features from EEG time series.

Main Results:

• The proposed model effectively learns correlations and temporal information across multiple EEG channels, improving emotion classification.
• Achieved high accuracy on the DEAP dataset: 95.02% for arousal and 94.61% for valence.
• Demonstrated strong performance on the SEED dataset, with an average emotion recognition accuracy of 95.49%.

Conclusions:

• The novel EEG emotion recognition method demonstrates significant potential for accurate mental state analysis.
• The combination of depthwise separable convolution and ON-LSTM offers an efficient and effective approach for EEG feature extraction.
• The model's high accuracy on public datasets validates its capability in real-world emotion recognition applications.