|
In order to analysis human pulse more effectively and design a more perfect neural network identifier than current ones in process of pulse patterns recognition, the physiological significance of human pulse,as an important reference of selecting pulse feature parameters used to identify pulse,is fully considered. Based on traditional Chinese medicine(TCM) pulse theory, the pulse time domain and frequency domain feature parameters are extracted, and then the correlation dimension, maximum Lyapunov exponent and Kolmogorov entropy, which are used as the chaos feature parameters of pulse and used to quantitatively verify that the pulse is a typical chaotic signal, are received by calculating in the reconstructed multidimensional phase space of pulse. Finally, the improved echo state network (ESN) identifier, whose activation function is switched to non-symmetric function combined with chaos theory, is designed and used to train and test 12 kinds of pulse patterns. And the main parameters of the novel neural network are optimized by particle swarm optimization (PSO) algorithm. Experiments show that the pulse feature parameters are selected effectively and the improved ESN neural network is more superior to feedforward neural networks, such as back propagation (BP) neural network, probabilistic neural network (PNN) and radial basis function (RBF) neural network. |
|
Keywords:Pulse patterns recognition; feature parameters extraction; chaotic features analysis; echo state network; PSO algorithm |
|