Algorithm 1. The description of DR-Net for EEG Depression Recognition

Require: EEG data $X$, labels $Y$, learning rate lr, batch size B, and training epoch $e$. Ensure: The model prediction $\widehat{Y}$. 1. Preprocess and segment EEG signals; 2. Augment EEG data through the GAN; 3. Calculate node features and adjacency matrix to generate the graph; 4. Construct the EEG dynamic graph and input it into the TPGCN to update the node features; 5. Convert the output of TPGCN to the data dimension; 6. Fuse the temporal dependencies of long-term time steps based on learned weights through the transformer module; 7. Input the long-term temporal features into the fully connected layer; 8. Perform gradient descent using the loss function; 9. Repeat steps 2–8 to train the model; 10. Return;