ACFSENet: an adaptive cross-frequency global sparse encoding network for end-to-end EEG emotion recognition

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• End-to-end Electroencephalography (EEG)-based emotion recognition is crucial for applications like human-computer interaction and affective brain-computer interfaces (aBCIs).
• Existing methods often neglect cross-frequency neural oscillation interactions and exhibit high computational complexity, hindering real-time and resource-constrained applications.

Purpose of the Study:

• To develop a novel end-to-end neural architecture, ACFSENet, for efficient and accurate EEG-based emotion recognition.
• To address the limitations of existing methods by integrating adaptive cross-frequency modeling and global sparse encoding.

Main Methods:

• ACFSENet utilizes an adaptive frequency-aware mechanism to dynamically focus on subject- and task-specific brain dynamics.
• A sparse attention mechanism with temporal distillation is incorporated to reduce computational complexity while maintaining long-range temporal dependency modeling.
• The model was evaluated using cross-block validation on the DEAP, SEED, and SEED-IV benchmark datasets.

Main Results:

• ACFSENet demonstrated superior performance compared to state-of-the-art methods in EEG-based emotion recognition.
• The proposed architecture achieved a significant balance between high recognition accuracy and computational efficiency.
• The adaptive frequency-aware mechanism enhanced the flexibility of emotional representation.

Conclusions:

• ACFSENet offers a promising solution for real-time and efficient EEG-based emotion recognition.
• The integration of adaptive cross-frequency modeling and sparse encoding effectively addresses the limitations of previous approaches.
• This work advances the development of affective computing and brain-computer interfaces.