The invention provides an
epilepsy signal classification method based on EWT and improved CSP. The method comprises the following steps: preprocessing the
signal, preferentially selecting a
signal channel by utilizing
correlation analysis, and extracting the characteristics of three characteristic domains, namely a
time domain, a
frequency domain and a space domain, of the epileptic electroencephalogram signal in an
attack interval and an
attack early stage; and finally, inputting the combined
feature matrix of the three feature domains into a classifier based on a
support vector machine to realize effective identification of the epileptic electroencephalogram signal
attack interval and the epileptic electroencephalogram signal attack earlier stage. The
time domain features comprise a meansquare value root, an absolute average value and a
zero crossing point number. According to the
frequency domain feature extraction method, empirical
wavelet transform is used for carrying out multi-mode
decomposition, a Welch power spectrum is used for carrying out single-
mode selection, and then a Hilbert transform method is used for extracting the
instantaneous amplitude and instantaneous frequency of a signal for the single mode of each channel; and an improved CSP
algorithm is used for extracting airspace features. According to the method, the limitation of an EMD method is overcome through empirical
wavelet transform, and in addition, the improved CSP
algorithm has high recognition rate and calculation efficiency.