VMD based wavelet hybrid denoising and improved FBCCA algorithm: a new technique for wearable SSVEP recognition

The brain-computer interface based on steady-state visual evoked potential (SSVEP) has gained increasing attention due to its non-invasiveness, low user training requirement, and high information transfer rate. In order to enhance the performance of SSVEP detection, we propose a denoising strategy combining variational mode decomposition (VMD) with wavelet fusion, along with an improved filter bank canonical correlation analysis (FBCCA) model for wearable SSVEP recognition. This denoising model employs detrended fluctuation analysis thresholding to identify noisy segments and applies deep filtering using discrete wavelet transform (DWT) or wavelet packet transform to the wearable data with significant noise. Finally, the FBCCA is used to classify the frequency bands segmented by the filter bank. Compared to single adaptive decomposition denoising and wavelet time–frequency denoising methods, our proposed approach focuses more on individual differences and achieves deep denoising effects through refined decomposition and adaptive IMFs selection. Experimental results demonstrate that compared to VMD denoising and adaptive wavelet denoising, our method improves classification accuracy by 0.34% and 2.59% for dry electrodes, and by 4.37% and 4.5% for wet electrodes. The hybrid model combining VMD decomposition with DWT denoising and enhanced FBCCA achieves optimal classification performance, providing a new perspective for wearable SSVEP recognition research and holds high potential for widespread application.