Research On Feature Extraction And Classification Recognition Of Pulse Waveforms

Traditional pulse diagnosis technology relies too much on the experience of physicians and lacks unified quantitative standards,so it is of great significance to use science and technology to modernize pulse diagnosis.The pulse waveform is collected by piezoelectric sensor,and the collected pulse waveform is preprocessed,feature extracted,and classified and identified.According to the shortcomings of the existing algorithms in these three parts of work,a series of improvements are proposed.In the preprocessing stage: the extreme-maximum method is used instead of the sliding window method to obtain the position of the starting point(minimum point).The position of the starting point is interpolated to obtain the envelope,and the baseline drift is removed by removing the envelope.This calculation method can identify the starting point according to the specific number of points contained in each cycle in the pulse waveform,and avoid the phenomenon of starting point recognition error caused by the fixed window of the sliding window method.In the feature extraction stage: according to the characteristic that the pulse waveform is a nonlinear complex signal,the approximate entropy,sample entropy,permutation entropy,fuzzy entropy,energy entropy,and power spectral entropy,etc.are extracted as nonlinear features,which enriches the characteristic information of the nonlinear domain of the pulse waveform.In the extraction algorithm of hidden tidal waves,the distance method is improved,and the original calculation method is changed.The experimental results show that the improved distance method can not only identify concealed tidal waves more accurately,but also identify different types of concealed tidal waves,and the overall calculation method is more universal.In the classification recognition stage,the Gini index of the random forest algorithm is used to calculate the importance of nonlinear features and verify the effectiveness of nonlinear feature introduction.The inertia weight of the basic PSO(Particle Swarm Optimization)algorithm is modified from constant type to exponential descent type,and the test function is used to verify and improve the optimization ability of the PSO algorithm.The basic PSO algorithm and the improved PSO algorithm are used to optimize the parameters of the RBF(Radial Basis Function)neural network,and the classification effect of the RBF neural network,the basic PSO-RBF neural network and the improved PSO-RBF neural network are compared.Experimental results show that the improved PSO-RBF neural network can effectively improve the accuracy of pulse classification,and this method has certain reference significance for the study of pulse species recognition system.