Pulse Signal Identification Based on Parameterized Bispectrum Estimation
Cai Kun-bao · Techniques of Automation and Applications · 2007
An AR model parameterized bispectrum estimation method is proposed to estimate the bispectra of the pulse signals for 15 heroin addicts and 15 healthy persons.The diagonal slice of the normalized bispectrum magnitude for every pulse wave is used to get 30 samples of the feature vector by using the K-L transform.The pulse signals are then identified by using these feature vectors and BP neural networks,the average identification rate of the trained networks reaches 96.7%.