Human Pulse Signal Preprocessing And Feature Extraction

Human pulse signal is one of the most important physiological signal, containing a large number of physiological and pathological information. Its changes and the occurrence of noise are often the earliest embodiment of human disease. Through human pulse investigation,we can know the physiological property of human body and supply important foundation for disease diagnosis. Signal preprocessing and feature extraction are researched in this thesis.Firstly, the current research status in the world is described, the formation mechanism,key properties and interference types of pulse signal are summarized. Then the advantages and disadvantages of wavelet transform and empirical mode decomposition(EMD) in the pulse signal denoising are analyzed. An improved algorithm of the combination to remove noise is proposed. Meanwhile the baseline drift is removed by using data fitting. The modified method is compared with wavelet and EMD methods. Experimental results show that under the same conditions the noise can be effectively eliminated in this algorithm, as well as the useful information and characteristic of the pulse signal are kept perfectly.Secondly, the feature extraction methods of pulse signal are researched. Pulse graph parameters are extracted, and the basic concept of K values is introduced. Then each mode of EMD decomposing is investigated. Modal energy quotient, Hilbert marginal spectrum and spectral energy ratio can be used in the analysis of the pulse signal. Finally, the normal pulse and atherosclerosis pulse are analyzed with the above methods. The experimental results show that the characteristic parameters obtained by these methods can effectively distinguish normal pulse and morbid pulse.