Data Augmentation Approach to Frequency Recognition of SSVEP Using Mask Encoding Combination Based Deep Learning

Steady state visual evoked potential (SSVEP) based brain computer interfaces (BCIs) are promising technique for real time communication and control. Utilizing transfer learning technique, this investigation introduces a novel classification method that incorporates mask encoding combination (MEC) data augmentation and convolutional neural networks (CNNs). The method’s superior classification performance is achieved by processing harmonics, channels, and temporal sub bands, which enhances the robustness of multi-channel EEG signal analysis. In a 1s time window, the approach obtains a maximal accuracy of 94.69% and a peak information transfer rate (ITR) of 193.14 bits min^−1 when evaluated on a benchmark dataset of 35 subjects and 40 characters. These findings surpass those of conventional methodologies, emphasizing the potential of integrating data augmentation and transfer learning to accelerate the development of SSVEP-based BCIs.