EEG-UDAF: A novel unsupervised domain adaptation framework leveraging dual-view frequency and spatio-temporal features for EEG-based emotion recognition

Currently, machine learning technology has been widely used in electroencephalogram (EEG) automatic emotion recognition tasks. However, the non-stationarity, individual differences, and high annotation costs in EEG signals seriously affect the generalization ability of the model. To solve these problems, we propose a new unsupervised domain adaptation framework named EEG-UDAF for EEG emotion recognition. First, we design a dual-view frequency domain feature fusion module that can extract rich frequency domain features from two perspectives, differential entropy and power spectral density, and fuse them based on the mutual cross-attention mechanism. Second, we propose a spatio-temporal feature extraction module based on graph convolutional network and bidirectional long short-term memory. It can comprehensively consider the spatial topological structure of the brain and the dynamic characteristics of temporal changes between EEG channels to extract complex spatio-temporal features. Third, we use the unsupervised domain adaptation strategy with a semantic correction mechanism to solve the non-stationarity and individual differences in EEG signals without manual annotation. We evaluate the EEG-UDAF on two public EEG emotion recognition datasets. Experiments based on intra-subject and inter-subject paradigms show that EEG-UDAF can effectively alleviate non-stationarity and individual differences.