Data Augmentation for EEG-based SSVEP Decoding

Steady-state visual evoked potential (SSVEP)based neural decoding (BCIs) play a crucial role in brain-computer interfaces (BCIs). However, traditional machine learning algorithms often face SSVEP decoding stability challenges due to the inherent multi-class small-sample pattern recognition, along with external factors such as noise interference or subject fatigue, resulting in unsatisfied performance. To solve this limitation, a data augmented decoding algorithm is proposed by applying Gaussian noise to SSVEP signals to enhance the robustness of traditional machine learning-based SSVEP decoding algorithms. This method was evaluated on selected subjects with initially low decoding accuracy in the public SSVEP dataset, showing significant performance improvement, with an average accuracy increase of 8.33%. The proposed approach effectively improves the robustness of SSVEP decoding with a straightforward manner and facilitates the broader application of SSVEP in practical BCIs.