CN107874750A

CN107874750A - Pulse frequency variability and the psychological pressure monitoring method and device of sleep quality fusion

Info

Publication number: CN107874750A
Application number: CN201711212983.5A
Authority: CN
Inventors: 邢晓芬; 陈光科; 江士尧; 林立韬; 陈东华
Original assignee: South China University of Technology SCUT
Current assignee: South China University of Technology SCUT
Priority date: 2017-11-28
Filing date: 2017-11-28
Publication date: 2018-04-06
Anticipated expiration: 2037-11-28
Also published as: CN107874750B

Abstract

The psychological pressure monitoring method and device merged the invention discloses a kind of pulse frequency variability and sleep quality, method are to gather human body photoelectricity volume pulse signal and acceleration signal first, identify current motion state.If recognition result is motion, give up this section of pulse signal data；If recognition result is gentle state, photoelectricity volume pulse signal is pre-processed first.Then, characteristic parameter extraction is carried out to pretreated pulse signal, extracted available for the multidimensional characteristic parameter for weighing human body psychological pressure.Then, the feature of extraction is inputted in the good grader of training in advance, output pressure state indices.Finally, with reference to nearest sleep quality, subsidiary classification device obtains final psychological pressure state.The present invention can reduce the cost and complexity of pressure monitoring, and the characteristic parameter for solving the problems, such as single physiological signal characterizes scarce capacity, while can greatly reduce the probability of miscarriage of justice that short distance monitoring is brought, and ensure the accuracy and feasibility of psychological pressure monitoring.

Description

Pulse rate variability and sleep quality fused psychological pressure monitoring method and device

Technical Field

The invention belongs to the technical field of mental health application, and particularly relates to a pulse rate variability and sleep quality fused psychological pressure monitoring method and device.

Background

Psychologists indicate that 15% -20% of the general population suffers from emotional disturbances and psychological distress. In fact, it is often seen in real life that many patients suffer from emotional excitement and overstress. Therefore, it is very important to correctly recognize the stress state of people in time in the study and life of people.

For the cognition of the self psychological pressure, the traditional method for evaluating the pressure by adopting a psychology related scale can accurately evaluate the self psychological pressure only by matching professional testers and correctly cognizing the testees, the actual operation is inconvenient, and the evaluation subjectivity is strong. Objectively, the psychological pressure of a person is also reflected in a physiological signal of the person, and some physiological parameter detection means based on electroencephalogram, electrocardio, electrodermal and the like can also objectively reflect the psychological pressure of the person, but the current psychological pressure monitoring technology based on the physiological signal has the following problems: (1) Or multiple special devices are adopted to obtain multiple physiological signals, so that the devices are expensive, the detection means are complex and difficult to popularize, or only one of electroencephalogram, electrocardio, skin electricity and other signals is collected, although the detection means is simplified, the problems of insufficient capability of characteristic parameters for representing the human body pressure state and low identification accuracy are caused. (2) The monitoring of the physiological signals of the human body is transient, the duration is several minutes or even shorter, and misjudgment is easily caused.

Objectively, the sleep quality and the psychological pressure of a person are closely related, and in papers 'research report on the relationship between pressure sources and phenomena of the pressure sources and the sleep quality', which are published by Yan Youwei, liu Mingyan and the like, under the background that the influence of different pressure sources and different pressure attributes on the sleep quality is more and more focused on the fields related to psychology and medicine, the influence of the different pressure sources and the different pressure attributes on the sleep quality is comprehensively explained once, so that the relationship between the psychological pressure of the human body and the sleep quality, namely the influence of the pressure on the sleep quality, and the sleep quality reflects the psychological pressure. In view of this, it is reasonable and feasible to use sleep quality to monitor human psychological stress.

In the prior art, in a patent with application publication number CN103584872a published on 2/19/2014, an applicant simultaneously acquires an electrocardiosignal, an electromyographic signal, a pulse signal and an electroencephalographic signal, integrates various physiological parameters, but has the following defects: (1) Various special devices (three patch electrodes for collecting electrocardiosignals, three patch electrodes for collecting electromyographic signals, three electrode plates for collecting electroencephalographic signals and a sensor for collecting pulse signals) are adopted to obtain various physiological signals, so that the devices are expensive, the detection means is complex and the popularization is difficult; (2) The monitoring of the human physiological signals is transient, the duration is about 1min, and misjudgment is easily caused.

In the prior art, liu Zhen in the article "psychological pressure identification research based on heart rate variability", an electrocardiographic signal acquisition device is used to acquire electrocardiographic signals of an experimental object, and then characteristic parameters are extracted from the electrocardiographic signals, which indeed simplifies the detection means compared with the aforementioned technology, but has the following defects: (1) Because only the electrocardiosignal, namely the physiological signal, is used, the capability of the characteristic parameter for representing the pressure state of the human body is insufficient, and the identification accuracy is low; (2) The monitoring of the human physiological signals is transient, the duration is about 2min, and misjudgment is easily caused.

Disclosure of Invention

The invention aims to overcome the defects and shortcomings of the prior art, and provides a pulse rate variability and sleep quality fused psychological pressure monitoring method which is simple and feasible, low in cost and strong in feasibility, and the pulse rate variability and the sleep quality are fused to jointly decide the human body pressure state, so that the defect of using a single physiological signal is overcome, and the probability of misjudgment can be greatly reduced.

Another object of the present invention is to provide a psychological stress monitoring device for implementing the above-mentioned psychological stress monitoring method.

The purpose of the invention is realized by the following technical scheme: the pulse rate variability and sleep quality fused psychological stress monitoring method comprises the following steps:

step1: collecting photoplethysmography signals and acceleration signals in the same time period, and judging the motion state of a human body by using the acceleration signals;

step2: if the motion state of the human body is judged to be a violent motion state, the photoelectric volume pulse signal of the section is abandoned, and if the motion state is judged to be a normal state, the photoelectric volume pulse signal is preprocessed;

and step3: extracting pulse rate variability characteristic parameters for measuring the human body pressure state from the preprocessed pulse signals;

and 4, step4: inputting the extracted characteristic parameters into a classifier which is trained off line and corresponds to a group to which the testee belongs, and pre-outputting a human body pressure state;

and 5: and combining the average sleep quality of N times of sleep closest to the monitoring time and the pre-output pressure state fusion decision of the classifier, and outputting the final human body pressure state.

Preferably, in step1, three-axis acceleration signals of the same time period are collected, and the motion state of the human body is judged by using the three-axis acceleration signals, wherein the method comprises the following steps:

setting the sampling frequency of a triaxial acceleration sensor as fs, the sampling time as Ts, and the total length of the sequence as N = Ts multiplied by fs; the amplitude sequence of the resultant acceleration of the three-axis acceleration is { a } ₁ ,a ₂ ,a ₃ ,…,a _N Therein of

Wherein a is _ix ,a _iy ,a _iz Respectively representing the component acceleration of the ith sampling point in the x, y and z directions; the time lengths of a front window and a rear window of the sliding time window are both T1, the sequence length is N1= T1 xfs, the time length of the middle window is T2, and the sequence length is N2= T2 xfs; a is _g Represents the acceleration of gravity;

(1-1) calculationWherein i =2 xn 1+ N2,2 xn 1+ N2+1, …, N;

(1-2) ifHas a _fi ＞a _{fThresholdValue} If the motion state is the violent motion state, otherwise, the motion state is determined as the normal state, wherein a _{fThresholdValue} Is a threshold value for threshold value judgment.

Preferably, in step2 of the present invention, the step of preprocessing the photoplethysmography signal is, in view of low time complexity, as follows:

setting the sampling frequency of the photoplethysmography sensor as Fs, the sampling time as Ts, and the total length of the pulse signal sequence as N' = Ts multiplied by Fs; pulse sequence is represented by x = { x = ×) ₁ ,x ₂ ,x ₃ ,…,x _N′ }; the upper limit cutoff frequency of the band-pass filtering is fpL, and the lower limit cutoff frequency is fpH;

(2-1) performing fast Fourier transform on the pulse signal sequence to obtain a real part sequence re = { re } of Fourier coefficients ₁ ,re ₂ ,…,re _N′ } and imaginary sequence lm = { lm = ₁ ,lm ₂ ,…,lm _N′ }；

(2-2) calculating the subscript range of frequency components to be retained:

due to the center symmetry of the spectrum, the subscript ranges areThe frequency components of (a) remain as well;

(2-3) zeroing frequencies outside the passband range:

(2-4) performing inverse Fourier transform to obtain pulse signal y = { y ] after band-pass filtering ₁ ,y ₂ ,y ₃ ,…,y _N′ }。

Preferably, in step3, the pulse rate variability feature parameter is extracted from the preprocessed pulse signal, and the step is:

(3-1) extracting the statistical characteristic parameters of the photoplethysmography sequence in the time domain;

(3-2) positioning the wave crest of the pulse signal, namely searching the wave crest P wave position of the pulse wave of the photoplethysmography;

(3-3) acquiring a PP interval, and acquiring a PP interval sequence according to the peak position and the sampling frequency;

(3-4) resampling the PP interval by adopting a cubic spline interpolation value to obtain a uniform sampling sequence;

(3-5) extracting the statistical characteristic parameters of the PP interval in the time domain;

(3-6) performing PP interval spectrum analysis, acquiring a power spectrogram of the resampled PP interval signal, and calculating to obtain a frequency domain characteristic parameter of the PP interval;

and (3-7) drawing a Poincare scatter diagram of the PP interval, and extracting the nonlinear characteristics of the PP interval.

Further, the statistical characteristic parameters in step (3-1) include: and the statistical average value of the mean value, the variance, the maximum value, the minimum value, the difference of the maximum value, the first-order difference and the second-order difference of the signal amplitude in each time window.

Furthermore, in the step (3-2), a maximum value method and a threshold value method are combined to carry out pulse signal peak positioning, and the steps are as follows:

let the peak position sequence of the pulse signal be index = { p ₁ ,p ₂ ,…,p _l Where l is the number of currently located peaks; threshold value is a threshold value set according to the normal range of the interval between two heartbeats of the human facies;

for y _i I =2,3,4, …, N' -1, if y _i ＞y _i-1 And y is _i ＞y _i+1 And is I is added to the sequence of peak positions.

Further, in step (3-4), resampling the PP interval by using cubic spline interpolation, comprising:

the PP interval sequence is Period = { Period = { (Period) ₁ ,Period ₂ ,…,Period _m Where m is the PP interval sequence length,

taking the sampling time of the (i + 1) th peak as the sampling time of the (i) th PP interval, the sampling time of the obtained PP interval sequence is T = { T = ₁ ,t ₂ ,…,t _m Therein of

T and Period are input into a cubic spline interpolation function, and an interpolation coefficient a3 of each segment can be obtained by a catch-up method _i ,a1 _i ,b3 _i ,b1 _i ,i＝1,2,…,m-1；

Let the interpolated PP interval sequence be Period '= { Period' ₁ ,Period′ ₂ ,…,Period′ _m-1 - }, wherein Period' _i ＝{Period′ _i1 ,Period′ _i2 ,…,Period′ _ik H, i =1,2, …, m-1 represents the i-th resampled PP interval sequence, and the sequence length is k;

let the resampling sampling interval of each segment be step, then

Period′ _ij ＝a3 _i ×(t _i+1 -(t _i +j×step)) ³ +a1 _i ×(t _i+1 -(t _i +j×step))+b3 _i

×((t _i +j×step)-t _i ) ³ +b1 _i ×((t _i +j×step)-t _i )

i＝1,2,…,m-1,j＝1,2,…,k

After interpolation, the non-uniformly sampled PP intervals are resampled to a uniform sequence with a sampling frequency of

Further, in step (3-5), extracting the statistical characteristic parameter of the PP interval in the time domain, including: the standard deviation of PP intervals SDNN, the root mean square of the difference between adjacent PP intervals RMSSD, the standard deviation of the difference between adjacent PP intervals SDSD, the number of adjacent PP intervals greater than 50ms NN50, NN50 values account for the percentage of the total number of PP intervals PNN50.

Furthermore, before the PP interval spectrum analysis in the step (3-6), whether the spectrum resolution of the PP interval is greater than a preset value is checked through fast Fourier transform, and if the spectrum resolution of the PP interval is not greater than the preset value, the spectrum resolution of the PP interval sequence is improved by adopting a method of zero padding at the end of the sequence. The method can reduce the requirement on the sampling frequency, and simultaneously avoid covering the low-frequency components of the PP interval caused by the over-short length of the PP interval sequence.

Furthermore, in step (3-6), the frequency domain characteristic parameters of the PP interval comprise: low-frequency component power LP, high-frequency component power HP, low-frequency power and high-frequency power ratio ratios of the interval PP; the frequency range of the characteristic parameter is illustrated as follows: f1=0.003hz, f2=0.04hz, f3=0.15hz, f4=0.4hz, f5=0.5hz; the interval frequency component range of human PP is f 1-f 5; the frequency range of the low-frequency component is f 2-f 3; the frequency range of the high-frequency component is f3 to f4.

Further, in the step (3-7), the non-linear characteristic of the PP interval comprises: poincare scatter diagram based on the vector angle index VAI and vector length index VLI of Poincare scatter diagram, the Poincare scatter diagram of PP intervals is implemented in a rectangular coordinate system by using PP _i As abscissa, PP _i+1 Making n-1 scatter points as ordinate, i =1,2, …, n-1;

whereinWherein

Preferably, the classifier in step4 is designed to perform a pressure excitation experiment to acquire physiological signals of an experimental object in a training stage, and the obtained training data covers data of a human body in a normal state and in different pressure states: in the experiment process, different pressure states are excited by controlling the time interval of the occurrence of the front digit and the back digit, and the experimental subjects of the pressure excitation experiment cover groups of different age groups; and further, pulse signals of groups of different ages are utilized, and characteristic parameters are extracted to train a customized classifier of the groups.

Preferably, in step 5, the quantitative index of the sleep quality is the proportion of deep sleep time to total sleep time, the sleep quality monitoring adopts a sleep quality monitoring method based on acceleration signals and wrist activity, the sleep quality value is output, and the steps are as follows:

unit ofThe time is recorded as T; the amplitude, the wrist activity and the waking state value of the triaxial composite acceleration signal in the ith unit time are respectively marked as a _i 、A _i 、D _i (ii) a Count values of unit time of awake state, light sleep and deep sleep are respectively recorded as c ₁ 、c ₂ 、c ₃ ；D _min 、D _max Lower and upper threshold values for the wake-up state values, respectively;

wrist activity is derived from the acceleration signal using proportional-integral:

calculating a wake-sleep state value:

where N5, N6 relate to a range of wrist activity levels associated with the wake state value for the ith unit of time, P _i+j Control A _i+j The degree of contribution to the determination of the sleep-wake state value;

if the waking state value D in the ith unit time _i Less than a set lower threshold D _min If yes, the unit time of the ith is judged to be in a deep sleep state, and the count value c of the deep sleep state ₃ Adding one; if D is _i Between a lower threshold D _min And an upper threshold D _max In between, the sleep state is judged to be in a light sleep state, and the count value c of the light sleep state ₂ Adding one; if D is _i Greater than a set upper threshold D _max If yes, the state is determined to be awake, and the awake state count value c ₁ Adding one;

calculating an indicator of sleep quality

Preferably, in step 5, the fusion decision step is: the four categories of the set classifier are normal state, low voltage state, medium voltage state,High pressure state, respectively using s ₁ ,s ₂ ,s ₃ ,s ₄ Is represented by ₁ ,α ₂ ,α ₃ ,α ₄ The posterior probabilities of the four states output by the classifier are respectively; the average sleep quality of the last three times is Q, and the lower threshold of the sleep quality is set to be Q _min The upper threshold is Q _max ；

Inputting: output of the classifier alpha ₁ ,α ₂ ,α ₃ ,α ₄ And the last three sleep quality averages Q;

output: fused post-decision pressure state s ₁ Or s ₂ Or s ₃ Or s ₄ ；

(5-1) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₁ Then, the pressure state after the fusion decision is judged as follows:

(5-2) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₂ Then, the pressure state after the fusion decision is judged as follows:

(5-3) if α ₁ ,α ₂ ,α ₃ ,α ₄ Middle maximum value of alpha ₃ Then, the pressure state after the fusion decision is judged as follows:

(5-4) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₄ Then, the pressure state after the fusion decision is judged as follows:

compared with the prior art, the invention has the following advantages and beneficial effects:

1. the invention adopts new wearable equipment to replace complex and expensive special equipment, uses the photoplethysmography technology to collect the pulse signals of the human body, extracts multidimensional pulse rate variability characteristic parameters which can effectively measure the pressure state of the human body after processing the pulse signals, and judges the real-time pressure state of the human body by combining a mode identification method.

2. Because the psychological pressure state of the human body and the quality of the sleep quality are closely related, the pulse rate variability and the sleep quality are fused to jointly decide the human body pressure state, the pulse rate variability and the sleep quality fusion not only can better represent the human body pressure state and solve the defect of using a single physiological signal, but also can greatly reduce the probability of misjudgment by using the sleep quality for the psychological pressure monitoring because the sleep quality monitoring is a long-term process.

3. The invention simultaneously carries out pulse sequence time domain analysis, PP interval frequency domain analysis and PP interval nonlinear analysis on the pulse signals to obtain characteristic parameters representing the human body pressure state in a multidimensional way, thereby improving the system identification rate.

4. The invention introduces a motion interference elimination mechanism to avoid the influence of violent motion on the characteristic parameters to cause misjudgment.

Drawings

FIG. 1 is a flowchart of the method of the present embodiment;

FIG. 2 is a flowchart illustrating the pulse rate variability feature parameter extraction in the method of this embodiment;

fig. 3 is a flowchart of a sleep quality monitoring method in the method of the present embodiment;

fig. 4 is a schematic diagram of fusion decision of pulse rate variability characteristic parameters and sleep quality in the method of the embodiment.

Detailed Description

The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.

Referring to fig. 1-4, this embodiment provides a specific pulse rate variability and sleep quality fused psychological stress monitoring method, which is detailed in the following steps.

Step1: collecting photoplethysmography signals and triaxial acceleration signals in the same time period, and judging the motion state of the human body by utilizing the triaxial acceleration signals.

Data acquisition: acquiring a photoplethysmography signal by using a photoelectric sensor on a smart watch, wherein the sampling frequency is Fs =25HZ, the sampling time is Ts =1min, and the sequence length is N' = Fs multiplied by Ts multiplied by 60=1500; the triaxial acceleration signal is acquired by using an acceleration sensor on a smart watch, the sampling frequency is fs =10HZ, the sampling time is also Ts =1min, and the sequence length is N = fs × Ts × 60=600.

Judging the motion state of the human body: adopting a sliding time window and a threshold value method, wherein the method comprises the following steps:

let the amplitude sequence of the resultant acceleration of the three-axis acceleration be { a } ₁ ,a ₂ ,a ₃ ,…,a _N Therein of Wherein a is _ix ,a _iy ,a _iz Respectively representing the component accelerations in the x, y and z directions of the ith sampling point.

The sliding time window consists of a front window, a middle window and a rear window, wherein the time length of the middle window is T2=2s, the sequence length is N2= T2 xfs =20, the time lengths of the front window and the rear window are both T1=3s, and the sequence length is N1= T1 xfs =30; a is _g Representing the acceleration of gravity, take a _g ＝9.8。

ComputingWherein i =2 xn 1+ N2,2 xn 1+ N2+1, …, N

If it isHas a _fi ＞a _{fThresholdValue} If the motion state is a violent motion state, otherwise, the motion state is determined as a normal state, wherein a _{fThresholdValue} Is a threshold value for threshold value judgment, and is determined by experiments, in the embodiment, a is taken _{fThresholdValue} ＝1.2。

When collecting sample data in a training stage, designing a pressure excitation experiment to collect physiological signals of an experimental object, and comprising the following steps of: adopting a mental calculation task experiment, and calculating the sum of numbers appearing on a screen at regular intervals by the experimental subject according to the requirement; in the pressure excitation experiment process, a multi-person competition form is adopted, and interference noise is added, so that the pressure of an experiment object is excited to the maximum extent.

The training data obtained according to the above experiment covers the data of the human body under normal state and different pressure states: in the experimental process, different pressure states are excited by controlling the time interval of digital occurrence before and after the experiment. The subjects of the stress-triggered experiments covered populations of different ages.

Step2: respectively carrying out different treatments according to different motion state discrimination results in the step1: if the motion state is identified as a violent motion state, the photoelectric volume pulse signal of the section is abandoned, and if the motion state is identified as a normal state, the photoelectric volume pulse signal is preprocessed.

Pulse signal preprocessing: the main purpose of the pulse signal preprocessing is to filter out irrelevant frequency components in the signal, and therefore, the original pulse signal is firstly passed through a band-pass filter.

In order to ensure the feasibility of the band-pass filtering algorithm on the wearable device, the invention uses an ideal band-pass filter and a fast Fourier algorithm with lower time complexity to reduce the time complexity to O (NlogN).

The frequency response of an ideal band-pass filter is as follows:

here, fpL and fpH are the lower limit cutoff frequency and the upper limit cutoff frequency of the ideal band pass filter, respectively, and since the number of normal heartbeats of the human body is 60 to 120 beats/s, in this embodiment, fpL =0.8HZ and fpH =2.2HZ are taken.

And (3) filtering by an ideal band-pass filter:

step1, performing fast Fourier transform on the pulse sequence to obtain a real part sequence re = { re } of Fourier coefficients ₁ ,re ₂ ,…,re _N′ } and imaginary sequence lm = { lm = ₁ ,lm ₂ ,…,lm _N′ }；

step 2. Calculate the subscript range of the frequency components to be retained:

due to spectral centrosymmetry, the subscript rangesThe frequency components of (a) remain as well;

step3, setting the frequencies outside the passband range to zero:

step4, performing inverse Fourier transform to obtain pulse signal y = { y ] after band-pass filtering ₁ ,y ₂ ,y ₃ ,…,y _N′ }。

Step3, extracting pulse rate variability characteristic parameters for measuring the human body pressure state from the preprocessed pulse signals, wherein as shown in fig. 3, the characteristic extraction process comprises the following steps: photoelectric volume pulse sequence time domain analysis, pulse signal peak positioning, PP interval acquisition, PP interval resampling, PP interval time domain analysis, PP interval frequency domain analysis and PP interval nonlinear analysis.

Photoelectric volume pulse sequence time domain analysis: the time domain analysis of the pulse sequence adopts a sliding time window method, firstly, the mean value, the variance, the maximum value, the minimum value, the difference between the most values, the first-order difference and the second-order difference value of the signal amplitude in each time window are extracted, and then the statistical average value is taken.

The sliding time window has a front window and a rear window both having a time length of T3=15s, a sequence length of N3= T3 × Fs =375, a middle window time length of T4=10s, and a sequence length of N4= T4 × Fs =250.

For the ith (i =2 × N3+ N4,2 × N3+ N4+1, …, N') time window:

mean value of pulse signal

Variance of pulse signal

Maximum value Max of pulse signal _i ＝max{y _i-N3-N4 ,y _i-N3-N4+1 ,…,y _i-N4 }

Minimum value Min of pulse signal _i ＝min{y _i-N3-N4 ,y _i-N3-N4+1 ,…,y _i-N4 }

Differ of the difference between the most significant pulse signals _i ＝Max _i -Min _i

First order difference of pulse signals

Second order difference of pulse signal

For the whole pulse signal sequence with the length of N':

mean value of pulse signal

Variance of pulse signal

Maximum value of pulse signal

Minimum value of pulse signal

First order difference of pulse signal

Second order difference of pulse signal

Positioning the wave crest of the pulse signal: the method for positioning the pulse wave crest by combining the maximum value method and the threshold value method is adopted, the influence of the pseudo wave crest is effectively eliminated, and the specific processes of the maximum value method and the threshold value method are as follows:

let the peak position sequence of the pulse signal be index = { p = ₁ ,p ₂ ,…,p ₁ Where l is the number of currently located peaks; threshold value is a threshold value set according to a normal range of two adjacent heartbeat intervals of a human body, and is set to be =0.45 in the embodiment.

For y _i (i =2,3,4, …, N' -1) if y _i ＞y _i-1 And y is _i ＞y _i+1 And is I is added to the sequence of peak positions.

Obtaining a PP interval: after the pulse signal wave crest is determined, the pulse beat interval PP interval can be obtained by combining the sampling frequency, which is concretely as follows:

let the PP interval sequence be Period = { Period = { [ Period } ₁ ,Period ₂ ,…,Period _m Where m +1 is the number of peaks.

And (3) PP interval resampling: resampling the PP intervals by adopting cubic spline interpolation to convert the PP intervals into a uniformly sampled sequence, wherein the specific process comprises the following steps:

T and Period are input into a cubic spline interpolation function, and an interpolation coefficient a3 of each segment can be obtained by a catch-up method _i ,a1 _i ,b3 _i ,b1 _i ,i＝1,2,…,m-1。

Let the interpolated PP interval sequence be Period '= { Period' ₁ ,Period′ ₂ ,…,Period′ _m-1 - }, wherein Period' _i ＝{Period′ _i1 ,Period′ _i2 ,…,Period′ _ik And j =1,2, …, and m-1 represents the i-th resampled PP interval sequence with a sequence length of k.

Let the resampling sampling interval of each segment be step, then

×((t _i +j×step)-t _i ) ³ +b1 _i ×((t _i +j×step)-t _i )

i＝1,2,…,m-1,j＝1,2,…,k

This embodiment takes step =0.04, resampling non-uniformly sampled PP interval sequences to a sampling frequency ofThe homogeneous sequence of (a).

And (3) PP interval time domain analysis: the method for extracting the statistical characteristic parameters of the PP interval in the time domain analysis process of the PP interval comprises the following steps: the standard deviation SDNN of the interval of PP, the root mean square RMSSD of the difference between the adjacent intervals of PP, the standard deviation SDSD of the difference between the adjacent intervals of PP, the percentage PNN50 of the number NN50 and NN50 values of the difference between the adjacent intervals of PP greater than 50ms in the total interval of PP, and the characteristic parameters are defined as follows:

the PP interval sequence after re-sampling is PP = { PP = { PP = ₁ ,PP ₂ ,…,PP _n }；

Wherein

Wherein Δ PP _i ＝PP _i+1 -PP _i

Wherein

Where NN50 is the number of adjacent PP intervals that differ by more than 50 ms.

And (3) PP interval frequency domain analysis:

before carrying out spectrum analysis on the PP interval sequence, firstly checking whether the spectrum resolution of the PP interval is large enough through fast Fourier transform, and if not, improving the spectrum resolution of the PP interval sequence by adopting a sequence end zero filling method; the method can reduce the requirement on the sampling frequency, and simultaneously avoid covering the low-frequency component of the PP interval caused by the over-short length of the PP interval sequence, and the specific process is as follows:

firstly, fast Fourier transform is carried out on the PP interval to obtain a Fourier coefficient XK = { XK = of the PP interval ₁ ,XK ₂ ,…,XK _n Where n is the length of the PP interval sequence after resampling, then its base frequency is

Setting fMin as the lowest frequency of the frequency spectrum components of the inter-heartbeat interval in the normal state of the human body, and generally setting fMin =0.04HZ;

if baseBand > fMin, the spectrum resolution of the PP interval is too small, and rho zeros are supplemented at the rear part of the PP interval sequence, so that a new fundamental frequency is obtained

In this embodiment, ρ =100 is taken to improve the spectral resolution of the PP interval to have fundamental frequencies of 0.02hz and (-fmin).

And (3) PP interval frequency domain analysis: extracting frequency domain parameters of the PP interval in the process of frequency domain analysis of the PP interval, wherein the process comprises the following steps: low-frequency component power LP, high-frequency component power HP, low-frequency power and high-frequency power ratio ratios of the interval PP; the frequency range of the characteristic parameter is illustrated as follows:

f1＝0.003HZ,f2＝0.04HZ,f3＝0.15HZ,f4＝0.4HZ,f5＝0.5HZ；

the interval frequency component range of human PP is f 1-f 5; the frequency range of the low-frequency component is f 2-f 3; the frequency range of the high-frequency component is f3 to f4.

And (3) nonlinear analysis of PP intervals:

drawing a Poincare scatter diagram of the PP interval in the nonlinear analysis process of the PP interval, and extracting the nonlinear characteristics of the PP interval, wherein the Poincare scatter diagram comprises the following steps: vector angle index VAI and vector length index VLI; poincare scatter plots and non-linear characteristic parameters for PP intervals are illustrated below:

poincare scatter diagram of PP interval is implemented by using PP in rectangular coordinate system _i As abscissa, PP _i+1 Making n-1 scatter points as ordinate (i =1,2, …, n-1);

wherein

Wherein

And 4, inputting the extracted characteristic parameters into an SVM classifier which is trained off-line and corresponds to the group to which the testee belongs, and pre-outputting the human body pressure state. In this embodiment, the classifier is an SVM classifier, and other classification networks such as threshold classification may also be used in other applications, which are not described in detail herein.

Training the SVM classifier: because the physiological parameters of the populations of different age groups have difference to a certain degree, in order to improve the accuracy of pressure state identification, the invention adopts a model customized training method aiming at the populations of specific age groups: and extracting characteristic parameters for training the customized SVM classifier of the population by using pulse signals of the population at different ages.

The age group in this example is divided as follows:

TABLE 1 division of age groups

Group numbering	Age group
		1	7-14
2	15-18
		3	19-23
4	24-30
		5	31-40
6	41-48
		7	49-55
8	56-65

And 5, combining the average sleep quality of the three times of sleep closest to the monitoring time and the pre-output pressure state fusion decision of the SVM, and outputting the final human body pressure state.

Monitoring sleep quality: the quantitative index of the sleep quality is the proportion of the deep sleep time to the total sleep time; the sleep quality is monitored by a monitoring method based on acceleration and wrist activity.

As shown in fig. 3, the sleep monitoring method specifically includes:

monitoring is carried out in unit time T =2min, a wake-up state value of each unit time is obtained, and states in the unit time are judged according to set wake-up state threshold values: awake state, light sleep, deep sleep. The ratio of the unit time number of deep sleep to the unit time number of deep sleep is the sleep quality.

Let the amplitude, wrist activity and waking-up state of the three-axis resultant acceleration signal in the ith unit time be recorded as a _i 、A _i 、D _i (ii) a Count values of unit time of awake state, light sleep and deep sleep are respectively recorded as c ₁ 、c ₂ 、c ₃ ；D _min ＝0.6、D _max =1 is a lower limit and an upper limit threshold of the awake state value, respectively.

step1, obtaining the wrist activity amount in the ith unit time from the acceleration signal by using proportional integral:

step2, calculating the waking state value in the ith unit time:

where N5 and N6 are related to the wrist activity level range associated with the waking state value of the i-th unit time, N5=5, N6=2,P is taken in this embodiment _N5+1+j Control A _i+j The contribution degree of the determination of the sleep-wake state value of the ith unit time is P, the length of the sequence is N5+ N6+1, and the invention takes P = {0.010,0.015,0.025,0.045,0.050,0.085,0.050,0.045}.

step3, if the i-th sleep state value D in the unit time _i Less than a set lower threshold D _min If yes, the unit time of the ith is judged to be in a deep sleep state, and a count value c of the deep sleep state ₃ Adding one; if D is _i Between a lower threshold D _min And an upper threshold D _max In between, the sleep state is judged to be in a light sleep state, and the count value c of the light sleep state ₂ Adding one; if D is _i Greater than a set upper threshold D _max If yes, the state is determined to be awake, and the awake state count value c ₁ And adding one. The following is expressed using the formula:

step4, calculating the proportion of the deep sleep time to the total sleep timeAs an indicator of sleep quality.

And (3) fusion decision: the human body psychological pressure monitoring method provided by the invention uses an algorithm of fusing pulse rate variability and sleep quality, wherein pulse rate variability parameters are used as the input of an SVM classifier, the SVM pre-outputs a human body pressure state and sleep quality fusion decision, and the final pressure state is output.

As shown in fig. 4, the algorithm for fusion decision is as follows:

the four categories of the SVM classifier are respectively a normal state, a low-voltage state, a medium-voltage state and a high-voltage state, and are respectively represented by s ₁ ,s ₂ ,s ₃ ,s ₄ Is represented by ₁ ,α ₂ ,α ₃ ,α ₄ Respectively SVM outputsA posterior probability of the four states of (a); the average sleep quality of the last three times of sleep is Q, and the lower limit threshold of the sleep quality is set to be Q _min =0.15, upper threshold Q _max ＝0.25。

Inputα ₁ ,α ₂ ,α ₃ ,α ₄ And the last three sleep quality averages Q;

output fused decided pressure state s ₁ Or s ₂ Or s ₃ Or s ₄ ；

(5-1) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₁ Namely, when the pressure state category pre-output by the classifier is a normal state: if the average sleep quality is smaller than the lower limit threshold, the state after the fusion decision is a medium-voltage state; if the average sleep quality is between the lower limit threshold and the upper limit threshold, the state after the fusion decision is a low-voltage state; and if the average sleep quality is greater than the upper limit threshold, the state after the fusion decision is a normal state. The following can be expressed using the formula:

(5-2) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₂ When the pressure state category pre-output by the classifier is a low-pressure state: if the average sleep quality is smaller than the lower limit threshold, the state after the fusion decision is a medium-voltage state; if the average sleep quality is between the lower limit threshold and the upper limit threshold, the state after the fusion decision is a medium-voltage state; and if the average sleep quality is greater than the upper limit threshold, the state after the fusion decision is a low-voltage state. The following can be expressed using the formula:

(5-3) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₃ I.e. pressure state of classifier pre-outputWhen the state type is a medium-voltage state: if the average sleep quality is smaller than the lower limit threshold, the state after the fusion decision is a high-voltage state; if the average sleep quality is between the lower limit threshold and the upper limit threshold, the state after the fusion decision is a medium-voltage state; and if the average sleep quality is greater than the upper limit threshold, the state after the fusion decision is a medium-voltage state. The following can be expressed using the formula:

(5-4) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₄ Namely when the pressure state type pre-output by the classifier is a high-pressure state: if the average sleep quality is smaller than the lower limit threshold, the state after the fusion decision is a high-voltage state; if the average sleep quality is between the lower limit threshold and the upper limit threshold, the state after the fusion decision is a high-voltage state; and if the average sleep quality is greater than the upper limit threshold, the state after the fusion decision is a medium-voltage state. The following can be expressed using the formula:

the techniques described herein may be implemented by various means. For example, these techniques may be implemented in hardware, firmware, software, or a combination thereof. For a hardware implementation, the processing modules may be implemented within one or more Application Specific Integrated Circuits (ASICs), digital Signal Processors (DSPs), programmable Logic Devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, micro-controllers, electronic devices, other electronic units designed to perform the functions described herein, or a combination thereof.

For a firmware and/or software implementation, the techniques may be implemented with modules (e.g., procedures, steps, flows, and so on) that perform the functions described herein. The firmware and/or software codes may be stored in a memory and executed by a processor. The memory may be implemented within the processor or external to the processor.

Those of ordinary skill in the art will understand that: all or part of the steps for realizing the method embodiments can be completed by hardware related to program instructions, the program can be stored in a computer readable storage medium, and the program executes the steps comprising the method embodiments when executed; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.

The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims

1. The pulse rate variability and sleep quality fused psychological pressure monitoring method is characterized by comprising the following steps of:

and step3: extracting pulse rate variability characteristic parameters for measuring the pressure state of the human body from the preprocessed pulse signals;

2. The mental stress monitoring method according to claim 1, wherein in step1, three-axis acceleration signals of the same time period are collected, and the motion state of the human body is judged by using the three-axis acceleration signals, and the method comprises the following steps:

setting the sampling frequency of a triaxial acceleration sensor as fs, the sampling time as Ts, and the total length of the sequence as N = Ts multiplied by fs; the amplitude sequence of the resultant acceleration of the three-axis acceleration is { a } ₁ ,a ₂ ,a ₃ ,…,a _N Therein ofWherein a is _ix ,a _iy ,a _iz Respectively representing the component acceleration of the ith sampling point in the x, y and z directions; the time lengths of a front window and a rear window of the sliding time window are both T1, the sequence length is N1= T1 xfs, the time length of the middle window is T2, and the sequence length is N2= T2 xfs; a is a _a Represents the acceleration of gravity;

(1-1) calculationWherein i =2 xn 1+ N2,2 xn 1+ N2+1, …, N;

(1-2) ifHas a _fi ＞a _{fThresholdValue} If the motion state is a violent motion state, otherwise, the motion state is determined as a normal state, wherein a _{fThresholdValue} Is a threshold value for threshold value determination.

3. The psychological stress monitoring method according to claim 1, wherein in step2, the step of preprocessing the photoplethysmographic pulse signal is:

setting the sampling frequency of the photoplethysmography sensor as Fs, the sampling time as Ts, and the total length of the pulse signal sequence as N' = Ts multiplied by Fs; pulse sequence is represented by x = { x = ×) ₁ ,x ₂ ,x ₃ ,…,x _N′ }; the upper limit cut-off frequency of the band-pass filtering is fpL, and the lower limit cut-off frequency is fpH;

(2-1) fast Fourier transform of pulse signal sequenceLeaf transformation to obtain real part sequence re = { re) of Fourier coefficient ₁ ,re ₂ ,…,re _N′ } and imaginary sequence lm = { lm = ₁ ,lm ₂ ,…,lm _N′ }；

(2-2) calculating the subscript range of frequency components to be retained:

(2-3) zeroing frequencies outside the passband range:

4. The mental stress monitoring method according to claim 1, wherein in step3, pulse rate variability characteristic parameters are extracted from the preprocessed pulse signals, and the steps are as follows:

(3-5) extracting a statistical characteristic parameter of the PP interval in a time domain;

5. Psychological stress monitoring method according to claim 4, wherein said statistical property parameters in step (3-1) comprise: the statistical average value of the mean value, the variance, the maximum value, the minimum value, the difference of the maximum value, the first-order difference and the second-order difference of the signal amplitude in each time window;

in the step (3-2), a method of combining a maximum value method and a threshold value method is adopted to carry out pulse signal peak positioning, and the steps are as follows:

for y _i I =2,3,4, …, N' -1, if y _i ＞y _i-1 And y is _i ＞y _i+1 And is Add i to the sequence of peak positions;

in the step (3-4), resampling the PP interval by adopting cubic spline interpolation, comprising the following steps:

the PP interval sequence is Period = { Period = { (Period) ₁ ,Period ₂ ,…,Period _m -where m is the PP interval sequence length,

Inputting T and Period into cubic spline interpolation function, and obtaining interpolation coefficient a3 of each segment by using pursuit method _i ,a1 _i ,b3 _i ,b1 _i ,i＝1,2,…,m-1；

let the resampling sampling interval of each segment be step, then

×((t _i +j×step)-t _i ) ³ +b1 _i ×((t _i +j×step)-t _i )

i＝1,2,…,m-1,j＝1,2,…,k

In the step (3-5), extracting the statistical characteristic parameters of the PP interval on the time domain, including: the standard deviation of PP intervals SDNN, the root mean square of the difference between adjacent PP intervals RMSSD, the standard deviation of the difference between adjacent PP intervals SDSD, the number of adjacent PP intervals greater than 50ms NN50, NN50 values account for the percentage of the total number of PP intervals PNN50.

6. Psychological stress monitoring method according to claim 4, wherein before the step (3-6) of the spectral analysis of the PP interval, it is checked whether the spectral resolution of the PP interval is greater than a predetermined value by fast Fourier transform, and if not, the spectral resolution of the PP interval sequence is increased by zero padding at the end of the sequence.

7. Psychological stress monitoring method according to claim 4 or 6, wherein in step (3-6), the frequency domain characteristic parameters of the PP interval comprise: low-frequency component power LP, high-frequency component power HP, low-frequency power and high-frequency power ratio ratios of the interval PP; the frequency range of the characteristic parameter is illustrated as follows: f1=0.003hz, f2=0.04hz, f3=0.15hz, f4=0.4hz, f5=0.5hz; the interval frequency component range of human PP is f 1-f 5; the frequency range of the low-frequency component is f 2-f 3; the frequency range of the high-frequency component is f 3-f 4;

in the step (3-7), the non-linear characteristics of the PP interval comprise: poincare scatter diagram based on the vector angle index VAI and vector length index VLI of Poincare scatter diagram, the Poincare scatter diagram of PP intervals is implemented in a rectangular coordinate system by using PP _i As abscissa, PP _i+1 Making n-1 scatter points as ordinate, i =1,2, …, n-1;

whereinWherein

8. The psychological stress monitoring method according to claim 1, wherein the classifier of step4 is designed to collect the physiological signal of the experimental subject during the training phase, and the obtained training data covers the data of the human body under normal state and different stress states: in the experiment process, different pressure states are excited by controlling the time interval of the occurrence of the front digit and the back digit, and the experimental subjects of the pressure excitation experiment cover groups of different age groups; and further, pulse signals of groups of different ages are utilized, and characteristic parameters are extracted to train a customized classifier of the groups.

9. The mental stress monitoring method according to claim 1, wherein in step 5, the quantitative index of sleep quality is a proportion of deep sleep time to total sleep time, the sleep quality monitoring adopts a sleep quality monitoring method based on an acceleration signal and wrist activity, and a sleep quality value is output, and the steps are as follows:

recording the unit time as T; the amplitude, the wrist activity and the waking state value of the triaxial composite acceleration signal in the ith unit time are respectively marked as a _i 、A _i 、D _i (ii) a Count values of unit time of awake state, light sleep and deep sleep are respectively recorded as c ₁ 、c ₂ 、c ₃ ；D _min 、D _max Lower and upper threshold values for the wake-up state values, respectively;

calculating a wake-sleep state value:

if the waking state value D in the ith unit time _i Less than a set lower threshold D _min If yes, the unit time of the ith is judged to be in a deep sleep state, and the count value c of the deep sleep state ₃ Adding one; if D is _i Between a lower threshold D _min And an upper threshold D _max In between, the sleep state is judged to be in a light sleep state, and the count value c of the light sleep state ₂ Adding one; if D is _i Greater than a set upper threshold D _max If yes, the system is judged to be in the awake state, and an awake state count value c ₁ Adding one;

calculating an indicator of sleep quality

In step 5, the fusion decision step is: the four categories of the classifier are set to be a normal state, a low-voltage state, a medium-voltage state and a high-voltage state respectively by s ₁ ,s ₂ ,s ₃ ,s ₄ Is represented by ₁ ,α ₂ ,α ₃ ,α ₄ The posterior probabilities of the four states output by the classifier are respectively; the average sleep quality of the last three times is Q, and the lower threshold of the sleep quality is set to be Q _min The upper threshold is Q _max ；

output: fused pressure states after decision ₁ Or s ₂ Or s ₃ Or s ₄ ；

(5-3) if α ₁ ,α ₂ ,α ₃ ,α ₄ Maximum value of alpha ₃ Then, the pressure state after the fusion decision is judged as follows:

10. a psychological stress monitoring device for carrying out the method of psychological stress monitoring according to any one of claims 1 to 9.