Research On Signal Processing Methods Of SSVEP In Brain Computer Interface

Brain Computer Interface(BCI) is a new interactive way independent of the conventional information channels of brain, which has high potential value on helping brain injury patients in the future. Currently, BCI especially those based on steady state visual evoked potential(SSVEP) attracts more and more attention because of the advantages of high information transmission rate and no training required. However, the BCI system based on SSVEP is very sensitive to stimulation, so the methods that how to process SSVEP signal effectively will directly affect the performance of BCI system. This paper mainly researches on the effects of different distances between stimulus and different numbers of stimulus on SSVEP signal processing based on spectral characteristics methods, wavelet transform, Hilbert-Huang transform(HHT) and canonical correlation analysis(CCA). The main work is as follows:Firstly, this paper designs a stimulator scheme based on LCD to quickly setting up the numbers of stimulus, the distance between stimulus and frequency of stimulus. On this basis, this paper researches on feature extraction of SSVEP based on spectral characteristics methods, such as FFT, Short-time Fourier Transform and AR mpdel.The results show that,, FFT and STFT are more suitable for analyzing the relationship between each stimulus comparing to the AR model.Then, this paper researches on feature extraction of SSVEP based on wavelet transform and HHT to analyze the effects of different numbers of stimulus. Considering shortcoming that Empirical Mode Decomposition(EMD) is easy rusult in model mixing, this paper presents an improved HHT analysis combined with wavelet packet. The results show that, the improved HHT analysis can achieve more enhanced performance than wavelet packet and HHT. In addition, the results also show that with the numbers of stimulus increasing, competitions between stimulis become more intense, and the suppression of the frequency of target stimulation is more obvious, which increases the difficulty of feature extraction.Finally, this paper researches on recognition of SSVEP based on CCA, in condition of different distances and different numbers of stimulus. Among the study, the factors that affect this two construct methods for reference signals are discussed by comparing the performances of CCA and Mset CCA. In order to exact useful ingredient of SSVEP, this paper proposes an improved CCA analysis combined with wavelet transform. On the one hand, the results show that the improved CCA analysis is a more enhanced method than CCA. On the other hand, the results also show that with the distances between stimulus diecreasing or the numbers of stimulus increasing, competitions between stimulis become more intense, and the suppression of the frequency of target stimulation is more obvious, which increases the difficulty of recognition.This paper finally compares the recognitions based FFT, wavelet packet, HHT, improved HHT, CCA and improved HHT methods. Results show that the improved CCA analysis can achieve best performance, and other methods are better than the FFT.