Research On Detection Method Of Abnormal Pulse Signal

In recent years,the progressive development of pulse diagnosis is gradually moving towards to objective study.The first step of the objectification is to acquire the pulse signal with the pulse acquisition device,then use pulse signal to do biomedical analysis.However,the pulse signal as a kind of dynamic physiological signal,the real-time acquisition environment is variability,noise interference from the external environment and the state of collectors should not be underestimated.The external noise is too large,the acquisition is position offset and gatherers to move or speak,are likely to affect the quality of the pulse signals with high noise,local mutation and abnormal baseline drift.These anomalies will cause some difficulties for the later studies.Thus eliminating local interference pulse signal is the key research contents to the objectification of pulse diagnosis.The main purpose of this project is to do the real-time quality analyze of the collected pulse signal on the basis of the existing pulse collection equipment.If detected abnormal pulse signal,then point out the cause of abnormal signal and recollection the pulse signal.Then,the abrupt change period of the pulse signal can be eliminated by noise detection.Finally,the pulse signals are normalized(including denoising,removing baseline drift,period segmentation,etc.).In the research of pulse signal quality classification,the time-domain feature,frequency-domain feature and entropy feature are used as the classification feature.The time-domain characteristics represent the waveform information of the pulse signal.The frequency domain characteristic represents the frequency domain distribution of the pulse signal and the noise,and the entropy characteristic represents the complexity of the pulse signal.With the extracted feature,the quality classification of pulse signals is carried out by using LIBSVM classifier,and time domain analysis is added based on LIBSVM classification results,and the precision is 100%.In order to ensure the real-time performance of pulse signal analysis and reduce the redundant features,the subject has also carried on the feature selection.Compared two different feature selection methods.Finally,five-dimensional representative features are selected for real-time analysis of pulse signal usability.In the local anomaly detection of pulse signals with better quality,a method based on area and curve smoothness is proposed to distinguish local mutations caused by noise and local mutations caused by disease,and we proposed a method based on Hilbert-Huang transform and ARMA model prediction to remove noise-induced mutations.In the study of pulse signal normalization,the direction of the pulse signal is adjusted based on the pulse peak and trough position.The method of cubic wave spline interpolation is used to fit the baseline drift in the pulse signal.A method of constructing similarity network is proposed to eliminate the abnormal single period of pulse signal.And finally the pulse signal is divided into the standard single-cycle set.