Optimal channel dynamic selection for Constructing lightweight Data EEG-based emotion recognition.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Current electroencephalogram (EEG) emotion recognition methods often rely on extensive data and numerous channels, increasing computational complexity.
• Improving accuracy typically involves larger datasets and more complex feature extraction, leading to significant time and resource consumption.

Purpose of the Study:

• To develop a more efficient and accurate method for EEG-based emotion recognition.
• To introduce a lightweight data approach for dynamic optimal channel selection using Ant Colony Optimization (ACO) combined with Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM).

Main Methods:

• EEG signals were transformed to the frequency domain using Fast Fourier Transform (FFT).
• Differential Entropy (DE) features were extracted from specific frequency bands.
• Ant Colony Optimization (ACO) was used to determine optimal electrode channels based on CNN-LSTM classification accuracy.
• Initial learning rate and batch size were optimized for data characteristics.

Main Results:

• The ACO-CNN-LSTM method achieved an average accuracy of 96.59% for three-class emotion recognition (positive, neutral, negative) on the SJTU Emotion EEG Dataset (SEED).
• Computational efficiency was increased by 15.85% compared to traditional CNN-LSTM methods.
• Accuracy remained above 90% even when data volume was reduced by 50%.

Conclusions:

• The proposed ACO-CNN-LSTM method offers a significant improvement in both accuracy and computational efficiency for EEG emotion recognition.
• This approach enables effective emotion recognition with lightweight data, reducing the demand for extensive computational resources.