EEG-Based Cross-Subject Emotion Recognition Using GADF and CADAN

In recent years, EEG emotion recognition based on deep learning has received great attention. However, most of the EEG signals which input into deep learning models are unidimensional time series, resulting the inability to utilize the superiority of deep learning in images classification. To deal with this problem, the paper uses Gramian Angular Difference Field (GADF) to transform unidimensional EEG signals into two-dimensional images. Based on transfer learning, we adopt Domain Adaptation Network with Channel Attention (CADAN) consisting of channel attention mechanism and domain adaptation network for cross-subject emotion recognition task. Then, by using the leave-one-out method, the two-dimensional grayscale images obtained by GADF are divided into source and target domains. Then, these images are input into the CADAN model for cross-subject experiment. The experiment achieves the average accuracy of 0.6898 on the SEED dataset, indicating that the GADF _ CADAN method can effectively perform cross-subject task.