关键词: 电位分类, 小波时频分析, 小波基函数, KS检验

Abstract: The separation of spikes is a key problem for invasive brain-computer interface. To deal with the similarity of spike temporal profile and frequency feature, a method was proposed to represent spike feature using wavelet analysis technique. First, wavelet functions, such as db, sym, bior, were used as base function to achieve high-dimension wavelet coefficient as spike feature. Next, in order to decrease the dimension of spike feature, Kolmogorov-Smirnov (KS) test was performed to select a few coefficients. After that, unsupervised K-means clustering was calculated to complete spike sorting. The experimental results show that, when the neural signal is at the noise level 0.05dB, 0.1dB, 0.15dB, sorting performance varies slightly while changing wavelet base functions. In all of these functions, sym5 wavelet outperforms the other five wavelet functions in terms of misclassified rate of spikes (between 1.21%～1.81%). Compared with Principal Component Analysis (PCA), the proposed method based on sym5 wavelet performs better even for the heavy noise spike data.

Key words: spike sorting, wavelet time-frequency analysis, wavelet base function, Kolmogorov-Smirnov (KS) test