Accelerating 3D Convolutional Neural Network with Channel Bottleneck Module for EEG-Based Emotion Recognition.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Emotion recognition using electroencephalogram (EEG) signals is a growing research area.
• Existing deep learning models for EEG emotion recognition vary widely in methods and input features, often leading to high computational costs.
• There is a need for efficient deep learning models that maintain high accuracy in EEG-based emotion recognition.

Purpose of the Study:

• To propose a novel 3D convolutional neural network with a channel bottleneck module (CNN-BN) for accelerated EEG-based emotion recognition.
• To reduce computational complexity without sacrificing classification accuracy.
• To effectively utilize spatiotemporal information from EEG signals.

Main Methods:

• Developed a 3D convolutional neural network with a channel bottleneck module (CNN-BN).
• Constructed a 3D spatiotemporal representation of EEG signals as input.
• Extracted spatiotemporal EEG features to leverage both spatial and temporal information.
• Evaluated the model on valence and arousal classification tasks using the DEAP dataset.

Main Results:

• Achieved high classification accuracy: 99.1% for valence and 99.5% for arousal on the DEAP dataset.
• Significantly reduced model parameters by 93.08% and FLOPs (Floating Point Operations) by 94.94%.
• The CNN-BN model demonstrated superior performance and parameter efficiency compared to state-of-the-art models.

Conclusions:

• The proposed CNN-BN model effectively extracts spatiotemporal EEG features for accurate emotion recognition.
• This model offers significant computational acceleration with minimal loss in classification performance.
• The CNN-BN model presents a promising solution for efficient and accurate EEG-based emotion recognition.