Multi-channel Steady-state visual evoked potential generation based on classifier-free conditional denoising diffusion probabilistic model

Brain computer interface (BCI) is an interactive way of communication between the brain and external devices such as computers, without the need for peripheral neural pathways and muscles. Steady-state visual evoked potential is a commonly used paradigm for brain-computer interfaces, but its data acquisition procedures are time-consuming and can easily cause subject fatigue. Therefore, it is necessary to explore generative data augmentation for improving the training of deep learning-based steady-state visual evoked potential (SSVEP) classification models. This study applied the classifier-free conditional denoising diffusion probabilistic model (cDDPM) to generate multi electrode channel SSVEP data and conducted generation experiments on a benchmark SSVEP dataset which contains 40 different categories. The quality of the generated data was evaluated through spectrum visualization and evaluation metrics Frechet inception distance (FID) and inception score (IS), and the data augmentation effect was verified by expanding the generated data to the trainset of the classification model.