Adaptive Multi-Domain Joint Decoding Algorithm for SSVEP Signal Classification

Shuting Jia, Rui Cao

Abstract: Steady-state visual evoked potentials (SSVEPs) based brain-computer interfaces(BCI) have emerged as a prominent research area owing to their advantageous signal-to-noise ratio and resolution. However, due to cost constraints in the acquisition, enhancing the decoding efficacy of BCIs with limited time windows and data volumes is imperative for the progressive industrialization of this technology. Approach. In this investigation, we introduce a adaptive multi-domain joint decoding algorithm tailored to address the SSVEP signal classification challenge. We propose a deep learning classification method for learning space-time-frequency tri-domain features, which facilitates learning more effective information in the feature extraction phase. Meanwhile, an adaptive regularization technique is proposed to reduce network overfitting while suppressing individual subject differences. Main results. Our model's performance was validated on 12-class and 40-class datasets. Experimental findings indicate that our model achieves accuracies of 99.00% and 90.33% under the 1s time window, surpassing those of comparable methods. Significance. The superior performance exhibited within short time windows enhances the potential applicability of our model to plug-and-play electroencephalogram(EEG) devices. Furthermore, the proposed regularization method provides new ideas for suppressing individual biometrics.

External IDs:dblp:conf/ijcnn/JiaC25