Improving Cross-Subject Emotion Recognition Performance with an Encoder-Decoder Structure.

Area of Science:

• Neuroscience
• Computer Science
• Affective Computing

Background:

• Emotion recognition using electroencephalography (EEG) is crucial for affective computing.
• Cross-subject emotion recognition models underperform due to individual EEG differences.
• Existing domain adaptation methods often neglect sample correspondence between source and target domains.

Purpose of the Study:

• To develop a novel domain adaptation technique to improve cross-subject emotion recognition.
• To address the performance degradation caused by inter-subject variability in EEG data.
• To leverage temporal alignment for more effective knowledge transfer in EEG-based emotion recognition.

Main Methods:

• An encoder-decoder architecture, termed the EEG converter, was designed.
• The EEG converter incorporates convolutional and max pooling layers in the encoder, and upsampling and convolutional layers in the decoder.
• Differential entropy features of EEG signals were transferred between subjects using the EEG converter with a time alignment condition.

Main Results:

• The EEG converter demonstrated significant improvements in cross-subject emotion recognition performance.
• The proposed method outperformed existing domain adaptation algorithms on SEED, SEED-IV, and SEEDV datasets.
• Effective transfer of EEG features was achieved by utilizing temporal correspondence.

Conclusions:

• The EEG converter offers a promising approach for enhancing cross-subject emotion recognition accuracy.
• Temporal alignment is a key factor in improving domain adaptation for EEG-based emotion recognition.
• This method effectively mitigates the impact of individual differences in EEG data for emotion recognition tasks.