Information Similarity Analysisbased Data Quality Evaluation For Real-time Heart Rate Signal Obtained From Smart Bands

With the rapid development of the current society,people’s health consciousness has gradually improved,and they have begun to pay attention to the monitoring of physical health in daily life.Many wearable devices have been shipped to the market.Among them,smart bands with health monitoring functions are favored by many consumers due to their advantages such as non-invasiveness,simple operation,and small size.The smart bands collect the human body’s physiological signals such as pulse waves,galvanic skin response,and temperature in real time,and implement the health monitoring functions by extracting and analyzing relevant characteristics from physiological signals.The pulse wave is currently one of the most commonly used and important monitoring signals in smart bands.From the pulse waves,real-time heartrate-variability(HRV)information can be obtained.HRV is an important indicator for evaluating the function of autonomic nervous system,which is obtained from real-time heart rate signals.For the existing smart bands in market,many health monitoring functions are obtained based on real-time heart rate or HRV indicators.However,due to the influence of the smart bands’ hardware equipment,processing algorithms,and external environment,the real-time heart rate signals extracted from the smart bands is not accurate,which further affects the accuracy of the monitoring indicators.Therefore,before launching the smart bands into market or developing a new health monitoring function,the data quality evaluation of real-time heart rate signal obtained from smart bands is an essential part.At present,for data quality evaluation methods of real-time heart rate signals obtained from smart bands,researches mainly focused on the pulse waveform characteristics,and the HRV correlation analysis using the ECG measured real-time heart rate signals,which can only qualitatively evaluate the signal quality and lack a quantitative evaluation method.Therefore,this thesis proposed a new method for evaluating the data quality of real-time heart rate signals obtained from smart bands based on information similarity analysis.The main research contents are as follows:(1)The real-time heart rate extracted from the synchronously collected ECG signal was taken as the standard reference.In order to obtain a more accurate ECG real-time heart rate,combining several commonly used real-time heart rate extraction algorithms,a new method of extracting the ECG real-time heart rate is proposed to deal with the influence of various noises.The verification results of two public data sets showed that the algorithm proposed in this thesis generally outperforms the seven commonly used extraction algorithms.(2)A stress-inducing experiment was designed,and the ECG signals and pulse waves of two smart bands from different brands were synchronously collected from 30 subjects in the states of deep breathing,resting and stress.Firstly,based on the new proposed ECG real-time heart rate extraction method,the standard real-time heart rate signal was obtained as the reference.Then,a new real-time heart rate signal quality evaluation method based on the similarity of fine feature patterns between different time series was proposed to quantitatively evaluate the signal quality of the real-time heart rate obtained using the smart band.By calculating the similarity index between the real-time heart rate obtained from smart bands and the standard reference,the signal quality of different bands in different states were compared and analyzed.(3)The signal quality assessment results were applied to the detection of human stress to improve the detection accuracy.The stress state detection was mainly realized by extracting the HRV indicators,and then inputting to the classification algorithm.The results showed that,compared with the original signals,the real-time heart rate signals filtered by the signal quality evaluation increased the accuracy of stress state detection by an average of 4%,and in four states(deep breathing 0.16 Hz,deep breathing 0.1Hz,resting and stress state)classification the accuracy reached 83.3%.