Pulse Signal Quality Assessment And Emotion Recognition Application

Pulse signal is a common physiological signal in the human body and contains abundant physiological and pathological information.It has been widely used in disease prevention,clinical monitoring and treatment.However,the pulse signal is easily disturbed by various factors in the collection process,which makes the quality of the pulse signal decrease or even submerge the useful signal,resulting in a false alarm in the clinical monitoring system,increasing the workload of the physician and influencing the clinical decision-making effect.Accurate and rapid assessment of signal quality has important research significance and application value for reducing false alarms,reducing physician workload,and improving clinical decision making efficiency.The current research on signal quality assessment technology mainly focuses on methods such as detection of interference segments.This article mainly focuses on signal quality features and classification algorithms.The paper extracts the feature matrix reflecting the signal quality information,constructs the optimal classification model,learning the mapping relationship between features and signal quality changes,and realizes the signal quality evaluation.The main research work of this paper is as follows:(1)The pulse signal generation mechanism and waveform characteristics are briefly described,and the characteristics of pulse signals and common interference types are introduced.A sound signal acquisition experiment scheme was designed and an experimental environment was set up.The right and left index finger pulse signal data with different emotional information and quality information from 53 subjects were successfully collected.(2)Using the method of manual quality level labeling,a single pulse wave is divided into good quality or poor quality based on the waveform integrity and the degree of interference.The good quality is marked as 1 and the poor quality is marked as 0.Take a 5s data sliding window to segment the pulse signal.The quality level of each pulse signal is calculated as the mean of the single-cycle pulse wave quality levels it contains.(3)We extracted 19 feature matrices from the linear and non-linear space that describe the pulse signal quality change information.It includes 13 time domain features,3 frequency domain feature,and 3 nonlinear features.The Mann-Whitney U non-parametric test method was used to analyze and test the differences of feature sequences under different qualities.Spearman correlation analysis method was used to test the correlation degree between feature sequences to optimize the classification model.The results show that compared with the time domain features,the frequency domain features and the nonlinear features show obvious differences,sensitivity and distinguishability under different signal qualities,and there is a high correlation between similar features.(4)A classification model learning framework based on grid parameter optimization algorithm,10-fold cross validation technique and support vector machine is designed and constructed.Based on the optimal parameters and optimal feature subsets,signal quality assessment models of a single feature and multiple features are established.A one-to-one identification of two signal qualities is achieved.The average classification accuracy of classification model of multi-feature training is 88.51%,which is higher than the classification accuracy of the classification model of single feature training,and the corresponding standard deviation is smaller.This shows that the classification model based on the optimal feature combination training in this thesis has a good classification performance and certain generalization ability.(5)The feature sequence that reflect the emotion information of pulse signal is extracted.Based on the grid parameter optimization algorithm,10-fold cross validation technique,and SVM algorithm,the original pulse data and good quality data which from algorithm evaluation were used to build classification models,and to identify emotional states.The results show that the classification accuracy of the classification model constructed by the data after signal quality assessment is 80.24%,and the classification accuracy of the classification model built without the signal quality evaluation is 77.66%.The classification accuracy,sensitivity,and specificity are both have seen an increase.Therefore,the signal quality assessment technology designed in this paper has practical application value and theoretical significance for further signal processing.