Domain adaptation spatial feature perception neural network for cross-subject EEG emotion recognition

Area of Science:

• Affective computing
• Neuroscience
• Machine learning

Background:

• Emotion recognition is vital in affective computing, with EEG-based deep learning showing promise.
• Current deep learning models struggle to integrate spatial activity and topology features from EEG signals.
• Addressing this gap is crucial for advancing EEG-based emotion recognition.

Purpose of the Study:

• To propose a novel deep learning network, DSP-EmotionNet, for cross-subject EEG emotion recognition.
• To enhance the simultaneous capture of spatial activity and topology features from EEG signals.
• To improve the accuracy and robustness of EEG-based emotion recognition models.

Main Methods:

• A spatial activity topological feature extractor module (SATFEM) was designed to extract both spatial activity and topology features.
• DSP-EmotionNet utilizes SATFEM as its core feature extractor.
• The model was evaluated on cross-subject EEG emotion recognition tasks.

Main Results:

• DSP-EmotionNet significantly improved accuracy in cross-subject EEG emotion recognition.
• The model achieved an average accuracy of 82.5% on the SEED dataset.
• The model achieved an average accuracy of 65.9% on the SEED-IV dataset, surpassing state-of-the-art methods.

Conclusions:

• DSP-EmotionNet effectively addresses the limitations of existing models in capturing complex EEG signal features.
• The proposed method demonstrates superior performance in cross-subject emotion recognition.
• This work advances the field of EEG-based emotion recognition with practical implications for affective computing.