Heart rate detection method based on wearable sensor

Abstract

The invention relates to the field of data processing, and particularly provides a wearable sensor-based heart rate detection method, which comprises the following steps of: calculating a first possibility that all data points in a respiration data set belong to noise points preliminarily; calculating a second possibility that all data points in the respiration data set belong to the Cheese-Scrake respiration model; wherein the Cheese-Scrake breathing model represents breathing characteristics of the heart failure patient during normal breathing; determining a final likelihood that the data point belongs to the noise point based on the first likelihood and the second likelihood; determining whether the data point is noise data based on the final likelihood; if the data points are noise data points, the noise data points are removed from the respiration data set, and the heart rate change condition of the heart failure patient is determined through the remaining data points. Specifically, the method can eliminate abnormal breathing and more accurately monitor the heart rate in real time.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a heart rate detection method based on a wearable sensor. Background Art

[0002] Real-time heart rate monitoring is of great significance for patients with heart failure (HF) because it can provide key health status information, help adjust treatment plans in time, and prevent the disease from worsening. Heart rate monitoring based on wearable sensors provides a continuous and non-invasive way of health management for patients with heart failure.

[0003] Generally, wearable smart watches often use PPG (pulse wave continuous blood pressure monitoring) technology to use optical sensors to measure changes in blood volume, thereby monitoring the heart rate of heart failure patients in real time.

[0004] However, the accuracy of PPG technology in real-time monitoring of heart rate in patients with heart failure is affected by factors such as the amplitude of exercise and improper wearing, resulting in inaccurate monitoring results. Summary of the invention

[0005] To solve the above problems, the present application provides a heart rate detection method based on a wearable sensor, which can eliminate abnormal breathing data (noise data) and perform more accurate real-time monitoring of the heart rate.

[0006] To solve the above problems, the present application provides a heart rate detection method based on a wearable sensor, comprising: Calculate the first probability that all data points in the respiratory data set are initially noise points; Calculating a second probability that all data points in the respiratory data set belong to a Che-Kessler respiratory model; wherein the Che-Kessler respiratory model represents the respiratory characteristics of a heart failure patient during normal breathing; Determine a final probability that the data point belongs to a noise point based on the first probability and the second probability; Determining whether the data point is noise data based on the final probability; If the data point is a noise data point, the noise data point is removed from the respiratory data set, and the remaining data points are used to determine the heart rate change of the heart failure patient.

[0007] The first possibility that all data points in the respiratory data set are initially noise points is calculated, including: Calculating a first possibility that the current data point in the respiratory data set preliminarily belongs to a noise point based on a difference in respiratory depth between the current data point and data points surrounding the current data point in the respiratory data set, thereby determining a first possibility that all data points in the respiratory data set preliminarily belong to noise points; The surrounding data points of the current data point include two data points on the left side of the current data point and two data points on the right side of the current data point.

[0008] The step of calculating the first possibility that the current data point in the respiratory data set is initially a noise point includes: The first possibility that the current data point in the respiratory data set is initially a noise point is calculated using the following formula: ; in, Indicates the first possibility that the data point O is initially a noise point, represents the breathing depth of data point O, represents the breathing depth of the two data points to the left of data point O, represents the breathing depth of the two data points to the right of data point O, Represents the maximum value function.

[0009] The step of calculating the second probability that all data points in the respiratory data set belong to the Che-Kessler respiratory model comprises: Determine a third possibility that any two data points in the respiratory data set are adjacent upper and lower vertices of a cosine function corresponding to a Che-Kessler respiratory model; Determine a cosine function corresponding to the Che-Kessler breathing model based on the breathing depth of the two data points with the third greatest probability and the time of the two data points with the third greatest probability; A second possibility that all data points in the respiratory data set belong to the Che-Kerk respiratory model is determined based on a cosine function corresponding to the Che-Kerk respiratory model.

[0010] The third possibility of determining that any two data points in the respiratory data set are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler respiratory model includes: Determine the preliminary possibility that the arbitrary two data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model based on the direction of the curvature radius of any two data points in the respiratory data set, the number of data points between the arbitrary two data points, the first possibility that the data points between the arbitrary two data points are initially noise points, the number of data points having the same value as the arbitrary two data points, and the first weight correction coefficient; Correcting the preliminary possibility using a second weight correction coefficient, thereby obtaining a third possibility that the arbitrary two data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model; Among them, the first weight correction coefficient is determined based on the direction of the curvature radius of the i-th data point between the any two data points; the second weight correction coefficient is determined based on the derivative value of the linear function of the time interval constructed by the any two data points.

[0011] Wherein, the method comprises: If the curvature radius of the i-th data point between any two data points is in the direction If , the first weight correction coefficient is 1, otherwise, the first weight correction coefficient is 0; If the derivative value of the linear function of the time interval constructed by any two data points is less than 0, the second weight correction coefficient is 1; if the derivative value of the linear function of the time interval constructed by any two data points is greater than or equal to 0, the second weight correction coefficient is 0.

[0012] Wherein, determining the preliminary possibility that the two arbitrary data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model includes: The preliminary probability is determined using the following formula: ; ; ; in, It indicates the preliminary possibility that any two data points OA and OB are adjacent upper and lower vertices of the cosine function corresponding to the Che-Schmidt breathing model. represents an exponential function with the natural constant e as the base, and max represents the maximum value. Indicates the direction of the curvature radius of the data point OA, Indicates the direction of the curvature radius of the data point OB, Indicates the number of data points with the same value as data point OA, Indicates the number of data points with the same value as data point OB. Indicates the first The data point is initially the first possibility of belonging to the noise point, m represents the number of data points between data points OA and OB, Indicates the first The first weight correction coefficient of the first possibility that the data point initially belongs to the noise point, It means that any two data points OA and OB are the corresponding upper and lower vertices of the cosine function under the reasonableness of the angle relationship. It indicates the reasonableness of the angle between any two data points OA and OB, which are the adjacent upper and lower vertices of the corresponding cosine function.

[0013] Wherein, determining the cosine function corresponding to the Che-Kessler breathing model based on the breathing depth of the two data points with the third greatest probability and the time of the two data points with the third greatest probability includes: Determine the breathing depth difference between the two data points with the third greatest probability as the amplitude of the cosine function, and determine the period of the cosine function using the time difference between the two data points with the third greatest probability; Constructing a reference formula using the amplitude of the cosine function, the period of the cosine function, the breathing depth of the two data points with the third greatest probability, and the time of the two data points with the third greatest probability; Based on the reference formula, a phase parameter and an offset parameter of a cosine function corresponding to the Che-Kessler breathing model are obtained; A cosine function corresponding to the Che-Kerk breathing model is constructed based on the phase parameter and the offset parameter.

[0014] Wherein, determining the second possibility that all data points in the respiratory data set belong to the Che-Kerk respiratory model based on the cosine function corresponding to the Che-Kerk respiratory model includes: Determining a fourth possibility that the current data point in the respiratory data set preliminarily belongs to the Che-Kessler respiratory model based on the cosine function corresponding to the Che-Kessler respiratory model, thereby determining a fourth possibility that all data points preliminarily belong to the Che-Kessler respiratory model; Determine the fourth possible correction coefficient based on the time corresponding to the current data point and the time corresponding to the current data point and the left and right adjacent data points; The second possibility is corrected based on the correction coefficient of the fourth possibility to obtain the second possibility that the current data point belongs to the Che-Kessler breathing model, thereby determining the second possibility that all data points belong to the Che-Kessler breathing model.

[0015] The cosine function corresponding to the Che-Kerk breathing model is constructed based on the phase parameter and the offset parameter, including: The cosine function corresponding to the Che-Schmidt breathing model is: ; in, Indicates the breathing depth of data point A The absolute value of the difference between the breathing depth at data point B and Indicates the time of data point A The time with data point B The absolute value of the time difference, T represents the time of any data point, M represents the breathing depth of the data point calculated by substituting the time of any data point into the cosine function corresponding to the Che-Kess breathing model, represents the phase parameter, Indicates the offset parameter.

[0016] Different from the prior art, the present application provides a heart rate detection method based on a wearable sensor, including calculating the first possibility that all data points in the respiratory data set are initially noise points; calculating the second possibility that all data points in the respiratory data set belong to the Che-Kess breathing model; wherein the Che-Kess breathing model characterizes the respiratory characteristics of heart failure patients during normal breathing; determining the final possibility that the data point belongs to the noise point based on the first possibility and the second possibility; determining whether the data point is noise data based on the final possibility; if the data point is a noise data point, removing the noise data point from the respiratory data set, and using the remaining data points to determine the heart rate changes of the heart failure patient. Specifically, this method can eliminate abnormal respiratory data (noise data) in the respiratory data set and monitor the heart rate more accurately in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which: Figure 1 A schematic diagram of a flow chart of a first embodiment of a heart rate detection method based on a wearable sensor according to the present invention; Figure 2 A scatter plot of data monitored over a preset period of time; Figure 3 for Figure 1 A schematic flow chart of an embodiment of step S12; Figure 4 A scatter plot of the time intervals of all data points between adjacent upper and lower vertices of the cosine function provided in this application; Figure 5 for Figure 3 Schematic diagram of a flow chart of an embodiment of step S22 in FIG. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0019] Generally, when PPG technology is used to monitor the heart rate of heart failure patients in real time, the monitoring results may be inaccurate due to the influence of movement amplitude, improper wearing, etc. Specifically, smart watches are generally used to monitor the heart rate of heart failure patients in real time. Specifically, the accelerometer in the smart watch is used to monitor the tiny movements of the wrist and monitor the depth of breathing. However, due to the movement amplitude and improper wearing, some noise data will appear, and these noise data will lead to inaccurate monitoring results. This application further determines the normal breathing data of heart failure patients through the distinguishing features between noise data and the normal breathing data of heart failure patients, and finally monitors the heart rate through respiratory sinus bradycardia (RSA) to determine the heart rate changes of heart failure patients.

[0020] For details, see Figure 1 , is a flow chart of a first embodiment of a heart rate detection method based on a wearable sensor of the present invention, the method comprising: Step S11: Calculate the first possibility that all data points in the respiratory data set are initially noise points.

[0021] First, the breathing data monitored by the accelerometer in the smart watch is divided into a group according to preset time, such as five minutes, ten minutes, etc., and the depth of the data monitored in each preset time (i.e., breathing depth) is plotted as follows Figure 2 The scatter plot shown. There may be different time intervals between two adjacent data points in the scatter plot, which is because the time intervals between pulse beats are different.

[0022] Each set of collected respiratory data is defined as a respiratory data set. In a respiratory data set, there are normal respiratory data collected from heart failure patients, as well as abnormal noise data caused by movement amplitude and improper wearing. Abnormal noise data will affect the accuracy of heart rate monitoring of heart failure patients, so the noise data in the respiratory data set needs to be removed.

[0023] In one embodiment, step S11 specifically includes: calculating the first possibility that the current data point in the respiratory data set is initially a noise point based on the difference in respiratory depth between the current data point and the surrounding data points of the current data point in the respiratory data set, thereby determining the first possibility that all data points in the respiratory data set are initially noise points. The surrounding data points of the current data point include two data points on the left side of the current data point and two data points on the right side of the current data point.

[0024] Specifically, since the characteristic of noise is that the breathing depth fluctuates with the surrounding data points, the first possibility that the data point is a noise point can be determined based on the degree of fluctuation between the data point and the surrounding data points. Taking data point O as an example, the breathing depth difference between data point O and the four data points around data point O (two on the left and two on the right) can be calculated.

[0025] In the embodiment of the present invention, the two data points on the left and the two data points on the right of the data point O are specifically analyzed. Of course, in other embodiments of the present invention, the four data points closest to the data point O may also be directly analyzed, and there is no limitation to this.

[0026] In a specific embodiment, the first possibility that the current data point in the respiratory data set is initially a noise point is calculated using the following formula: ; in, Indicates the first possibility that the data point O is initially a noise point, represents the breathing depth of data point O, represents the breathing depth of the two data points to the left of data point O, represents the breathing depth of the two data points to the right of data point O, Represents the maximum value function.

[0027] For data point O, the greater the fluctuation of its breathing depth with the surrounding data points, the more likely it is to be a noise point, showing a positive proportional relationship. Because there may be noise points around the normal heart failure data points, affecting its fluctuation degree, and noise points cannot appear continuously around the normal heart failure data points, two data points on the left and right sides are selected here, the maximum difference is removed and the average is calculated, so as to calculate the fluctuation degree of data point O and the surrounding pixel points. This fluctuation degree is used to characterize the first possibility that data point O is initially a noise point.

[0028] Step S12: Calculate the second possibility that all data points in the respiratory data set belong to the Che-Kelly respiratory model; wherein the Che-Kelly respiratory model characterizes the respiratory characteristics of heart failure patients during normal breathing.

[0029] Specifically, the breathing characteristics of heart failure patients during normal breathing are in the Che-Kessler breathing model. In this breathing pattern, the breathing depth also fluctuates, so the fluctuation of breathing depth cannot be used as the basis for judging noise data. Since the fluctuation of breathing depth is not only a characteristic of noise data, but also a characteristic of the Che-Kessler breathing pattern, the Che-Kessler breathing model should be analyzed in this step. In the Che-Kessler breathing model, as the breathing frequency increases, the breathing depth first becomes deeper and then shallower, then stagnates for a period of time, then becomes deeper and shallower again, which is specifically manifested as the characteristics of the periodic pattern of the cosine function. Therefore, the second possibility of all data points belonging to the Che-Kessler breathing model can be calculated, and the first possibility of the initial noise data point can be weighted corrected with this second possibility to obtain the possibility that all data points ultimately belong to the noise data point.

[0030] Please combine the specific Figure 3, step S12 specifically includes: Step S21: Determine the third possibility that any two data points in the respiratory data set are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler respiratory model.

[0031] First, since the Che-Kessler breathing model shows a periodic pattern of a cosine function in which the breathing depth first becomes deeper and then becomes shallower, then stagnates for a period of time, and then first becomes deeper and then becomes shallower and pauses for a period of time, we can first calculate the second possibility by finding the cosine function of the Che-Kessler breathing model.

[0032] Specifically, based on the direction of the curvature radius of any two data points in the respiratory data set, the number of data points between any two data points, the first possibility that the data points between any two data points are noise points, the number of data points with the same numerical values ​​as any two data points, and the first weight correction coefficient, the preliminary possibility that any two data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler respiratory model is determined.

[0033] For all data points on the respiratory data, select the data points closest to the 90-degree and 270-degree directions in the direction of the curvature radius of all data points, and the curvature radius direction of all data points between these two vertices is between these two points, and the respiratory frequency is also increasing. To calculate the curvature of each data point, you can first perform a quadratic fit of the curve between each data point and the two adjacent data points on the left and right to obtain the curvature radius direction of all data points.

[0034] In a specific embodiment, the following formula is used to determine the preliminary possibility that the arbitrary two data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model: ; ; ; in, It indicates the preliminary possibility that any two data points OA and OB are adjacent upper and lower vertices of the cosine function corresponding to the Che-Schmidt breathing model. represents an exponential function with the natural constant e as the base, and max represents the maximum value. Indicates the direction of the curvature radius of the data point OA, Indicates the direction of the curvature radius of the data point OB, Indicates the number of data points with the same value as data point OA, Indicates the number of data points with the same value as data point OB. represents the first possibility that the jth data point between data points OA and OB is a noise point, m represents the number of data points between data points OA and OB, The first weight correction coefficient representing the first possibility that the jth data point between data points OA and OB is initially a noise point, It means that any two data points OA and OB are the corresponding upper and lower vertices of the cosine function under the reasonableness of the angle relationship. It indicates the reasonableness of the angle between any two data points OA and OB, which are the adjacent upper and lower vertices of the corresponding cosine function.

[0035] It should be noted that, in the above formula, the first weight correction coefficient is determined based on the curvature radius direction of the jth data point between the two arbitrary data points. Specifically, if the curvature radius direction of the jth data point between the two arbitrary data points, such as data points OA and OB, is If , the first weight correction coefficient is 1, otherwise the first weight correction coefficient is 0. The formula is: ; It is expressed as the direction of the curvature radius of the jth data point between data points OA and OB.

[0036] For the adjacent upper and lower vertices of the cosine function, one radius of curvature approaches 90° and the other radius of curvature approaches 270°, so and The difference and and The closer the maximum absolute value of the difference is to 0, the greater the initial possibility that it belongs to the adjacent upper and lower vertices, showing an inverse proportional relationship. For the cosine function, the upper and lower vertices are corresponding, and the number of upper vertices should be the same as the number of lower vertices, and the difference is small. Therefore, the smaller the difference in the number of data points with the same value as data points OA and OB on the time axis, the greater the possibility that they belong to the adjacent upper and lower vertices, showing an inverse proportional relationship. In addition, for all data points between OA and OB, the possibility of expressing suspected noise should be greater, that is, The larger the value, the higher the curvature. Because we need to select adjacent upper and lower vertices, and the curvatures of the upper and lower vertices are between these two curvatures, there are two possible points that are definitely not within these two curvatures. One is that the interval contains multiple cycles, which leads to it not being within the interval. The other means that if there are abnormal points, it may also lead to it not being within the interval of the angle where the curvature radius is located. Only the points between the curvatures of OA and OB are selected to construct The formula can then be used to calculate the average value, that is, , the larger the average value, the greater the possibility that it belongs to the adjacent upper and lower vertices. There is a positive proportional relationship.

[0037] Specifically, for the adjacent upper and lower vertices OA and OB of the cosine function, as the Che-Kess breathing model changes over time, the breathing frequency should increase. Therefore, a scatter plot can be made based on the breathing interval from left to right, and the time difference between each data point and the next data point is recorded as the time interval of each data point. The time interval curve of all data points between OA and OB is obtained as follows: Figure 4 The time interval scatter plot of all data points between adjacent upper and lower vertices of the cosine function is shown. The derivatives of these functions after linear fitting can be used to determine whether the overall respiratory rate between OA and OB points increases or decreases.

[0038] The following formula can be constructed, and a weight correction can be made to the preliminary possibility that the data points OA and OB belong to the adjacent upper and lower vertices of the cosine function of the Che-Ke breathing model by increasing and decreasing the breathing frequency. Specifically, the preliminary possibility is corrected using the second weight correction coefficient to obtain the third possibility that any two data points are the adjacent upper and lower vertices of the cosine function corresponding to the Che-Ke breathing model. The second weight correction coefficient is determined based on the derivative value of the linear function of the time interval constructed by any two data points. In a specific embodiment, if the derivative value of the linear function of the time interval constructed by any two data points is less than 0, then the second weight correction coefficient is 1, and if the derivative value of the linear function of the time interval constructed by any two data points is greater than or equal to 0, then the second weight correction coefficient is 0. The formula of the second weight correction coefficient is expressed as: ; in, represents the second weight correction coefficient, Represents the derivative value of the linear function of the time interval constructed by the data points OA and OB.

[0039] After determining the second weight correction coefficient, the second weight correction coefficient is used to correct the preliminary possibility, thereby obtaining the third possibility that any two data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kess breathing model. Since the respiratory frequency in the cosine function constructed by the Che-Kess breathing model increases between adjacent upper and lower vertices, a linear function can be constructed based on the time interval of adjacent data between data points. If the derivative of the linear function is less than 0, it proves that the respiratory frequency is increasing. If the derivative of the linear function is greater than or equal to 0, it proves that the respiratory frequency is decreasing. In this way, a weight correction is made to the possibility that the data points OA and OB belong to the adjacent upper and lower vertices of the cosine function of the Che-Kess breathing model. The calculation formula for the third possibility can be constructed as follows: .

[0040] in: Indicates that the data points OA and OB are the third possibility of the adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model, It indicates that the data points OA and OB are the preliminary possibilities of the adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model.

[0041] Step S22: Determine the cosine function corresponding to the Che-Kessler breathing model based on the breathing depths of the two data points with the third greatest probability and the times of the two data points with the third greatest probability.

[0042] Combination Figure 5 , step S22 specifically includes: Step S41: Determine the breathing depth difference between the two data points with the third greatest probability as the amplitude of the cosine function, and determine the period of the cosine function using the time difference between the two data points with the third greatest probability.

[0043] Specifically, the two data points with the third largest probability are recorded as points A and B, and the breathing depth of data point A is expressed as , the breathing depth of data point B is expressed as , the breathing depth is different That is, the amplitude of the cosine function, the time difference between data point A and data point B That is one quarter of the period of the cosine function, which is used to determine the amplitude and period of the cosine function.

[0044] Step S42: construct a reference formula using the amplitude of the cosine function, the period of the cosine function, the breathing depth of the two data points with the third greatest probability, and the time of the two data points with the third greatest probability.

[0045] Specifically, , Substituting into the cosine function formula, α represents the amplitude of the cosine function, and β represents the period of the cosine function, that is, the following reference formula can be constructed by combining them: Step S43: Obtaining the phase parameter and offset parameter of the cosine function corresponding to the Che-Kessler breathing model based on the reference formula.

[0046] By calculating the above reference formula, the phase parameter of the cosine function corresponding to the Che-Kessler breathing model can be obtained: , offset parameter .

[0047] Step S44: constructing a cosine function corresponding to the Che-Kessler breathing model based on the phase parameter and the offset parameter.

[0048] Specifically, the phase parameter , offset parameter Substituting into the cosine function, the cosine function corresponding to the Che-Kessler breathing model can be constructed as: ; in, Indicates the breathing depth of data point A The absolute value of the difference between the breathing depth at data point B and Indicates the time of data point A The time with data point B The absolute value of the time difference, T represents the time of any data point, M represents the breathing depth of the data point calculated by substituting the time of any data point into the cosine function corresponding to the Che-Kess breathing model, α represents the amplitude of the cosine function, and β represents the period of the cosine function.

[0049] Step S23: Determine the second possibility that all data points in the respiratory data set belong to the Che-Kerk respiratory model based on the cosine function corresponding to the Che-Kerk respiratory model.

[0050] In a specific embodiment, the fourth possibility that the current data point in the respiratory data set preliminarily belongs to the Che-Kerk breathing model is determined based on the cosine function corresponding to the Che-Kerk breathing model, thereby determining the fourth possibility that all data points preliminarily belong to the Che-Kerk breathing model.

[0051] Specifically, for data point O, the time To of data point O is substituted into the cosine function corresponding to the above-mentioned Che-Schmidt breathing model, and T in the cosine function is replaced to calculate the breathing depth M of data point O: To .

[0052] The breathing depth M based on the calculated data point O To and the breathing depth of data point O in the breathing data set The calculation data point O preliminarily belongs to the fourth possibility of the Che-Kessler breathing model. In one embodiment, the calculation formula of the fourth possibility is: ; in, Represents data points The fourth possibility that initially belongs to the Che-Schmidt breathing model, represents a hyperparameter so that the denominator is not 0, for example It can be 0.01. In the formula, It represents the fitting error between the data point O and the cosine function. The larger the error, the lower the possibility that the data point belongs to the Che-Kessler breathing model, showing an inverse proportional relationship.

[0053] Further, based on the characteristic of increased respiratory frequency in the Che-Kessler respiratory model, the correction coefficient of the fourth possibility is determined based on the time corresponding to the current data point and the time corresponding to the current data point and the left and right adjacent data points. Specifically, the correction coefficient formula of the fourth possibility is expressed as: ; in, Indicates the fourth possible correction coefficient corresponding to data point O. represents the time corresponding to the data point O, represents the time corresponding to the adjacent data point on the left side of data point O, Indicates the time corresponding to the adjacent data point on the right side of data point O.

[0054] The second possibility is corrected based on the correction coefficient of the fourth possibility to obtain the second possibility that the current data point belongs to the Che-Kessler breathing model, thereby determining the second possibility that all data points belong to the Che-Kessler breathing model.

[0055] In a specific embodiment, the second probability calculation formula of the current data point belonging to the Che-Kessler breathing model is: ;in, Represents data points The second possibility for the data point belonging to the Che-Schmidt breathing model, Represents data points The fourth possibility that belongs to the Che-Schmidt breathing model data point is Represents a correction coefficient for the fourth possibility.

[0056] because It only reflects whether the data point O is on the cosine function of the Che-Kessler breathing model, but cannot reflect whether it conforms to the characteristics of the accelerated breathing rate of the Che-Kessler breathing model. Make a correction to see if the breathing rate is accelerated. If it is accelerated, it may belong to the Che-Kessler breathing model.

[0057] Step S13: Determine the final possibility that the data point belongs to a noise point based on the first possibility and the second possibility.

[0058] Specifically, the second probability Q of all data points O belonging to the Che-Kessler breathing model is normalized: .

[0059] Using the second possibility after normalization The final possibility that the data point belongs to a noise point is determined by the first possibility. Specifically: ; in, Indicates the final possibility that data point O belongs to a noise point, Represents data points The first possibility belongs to the noisy data. is the second possibility that the data point O belongs to the Che-Schmidt breathing model data point, Represents the absolute value of the logarithmic function with base e.

[0060] Since the data point O is more likely to belong to the Che-Kess breathing model, its weight as a noise point should be lower, and the less likely it is to belong to the Che-Kess breathing model, its weight as a noise point should be higher. The function performs an expansion and differentiation process on this result. The closer the function is to 1, the closer the weight is to 0; the closer it is to 0, the greater the weight.

[0061] For normal heart failure patients, breathing is relatively stable, and the depth of breathing is consistent, showing a Che-Ke breathing pattern. In this pattern, the breathing rate increases, the breathing depth first becomes deeper and then shallower, then stagnates for a period of time, and then becomes deeper, then shallower, and then pauses for a period of time. Based on these characteristics, we can first exclude the influence of noise data caused by movement amplitude and improper wearing, and then monitor the heart rate through respiratory sinus bradycardia (RSA) through the normal breathing data set of heart failure patients, so as to perform more accurate real-time heart rate monitoring of heart failure patients.

[0062] Step S14: Determine whether the data point is noise data based on the final probability.

[0063] For the final possibility that all the above data points belong to noise, the maximum and minimum values ​​are first normalized, and then a threshold is set. This application takes a threshold of 0.8 as an example. If it exceeds this threshold, it is considered to be noise data.

[0064] Step S15: If the data point is a noise data point, the noise data point is removed from the respiratory data set, and the remaining data points are used to determine the heart rate change of the heart failure patient.

[0065] The noise data caused by the movement amplitude and improper wearing have been determined. By removing these noise data, a respiratory data set showing normal heart failure in each group of heart failure patients within five minutes can be obtained.

[0066] Further, the heart rate of heart failure patients can be more accurately monitored through respiratory sinus bradycardia (RSA) through the normal breathing data set of heart failure patients. Through real-time heart rate monitoring, the changes in heart rate can be observed. An increase in heart rate may indicate an increased heart load or a problem with the cardiovascular system, while a decrease in heart rate may be related to decreased heart function. By monitoring the trend of heart rate changes, the health of heart failure patients can be preliminarily judged.

[0067] The method of the present application performs real-time heart rate monitoring by determining a respiratory data set of a heart failure patient that shows normal heart failure. This technology can avoid the influence of noise data caused by the amplitude of movement and improper wearing in the PPG technology monitoring of heart rate, making real-time monitoring more accurate. Because the breathing depth of a normal heart failure patient is either a certain or a Che-Sk breathing pattern, the breathing frequency increases, and the depth of breathing presents a cyclical pattern of first deep, then shallow, and then pausing for a period of time. Based on these characteristics, the noise data is removed and normal breathing data is obtained to implement more effective monitoring of the heart rate of heart failure patients.

[0068] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0069] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A heart rate detection method based on a wearable sensor, characterized in that: include: Calculate the first probability that all data points in the respiratory data set are initially noise points; Calculating a second probability that all data points in the respiratory data set belong to a Che-Kessler respiratory model; wherein the Che-Kessler respiratory model characterizes the respiratory characteristics of a heart failure patient during normal breathing; Determine a final probability that the data point belongs to a noise point based on the first probability and the second probability; Determining whether the data point is noise data based on the final probability; If the data point is a noise data point, the noise data point is removed from the respiratory data set, and the remaining data points are used to determine the heart rate change of the heart failure patient.

2. A heart rate detection method based on a wearable sensor according to claim 1, characterized in that: Calculate the first possibility that all data points in the respiratory data set are initially noise points, including: Calculating a first possibility that the current data point in the respiratory data set preliminarily belongs to a noise point based on a difference in respiratory depth between the current data point and data points surrounding the current data point in the respiratory data set, thereby determining a first possibility that all data points in the respiratory data set preliminarily belong to noise points; The surrounding data points of the current data point include two data points on the left side of the current data point and two data points on the right side of the current data point.

3. A heart rate detection method based on a wearable sensor according to claim 2, characterized in that: Calculating a first possibility that a current data point in the respiratory data set is initially a noise point comprises: The first possibility that the current data point in the respiratory data set is initially a noise point is calculated using the following formula: ； in, Indicates the first possibility that the data point O is initially a noise point, represents the breathing depth of data point O, represents the breathing depth of the two data points to the left of data point O, represents the breathing depth of the two data points to the right of data point O, Represents the maximum value function.

4. A heart rate detection method based on a wearable sensor according to claim 1, characterized in that: Calculating a second likelihood that all data points in the respiratory data set belong to the Che-Kessler respiratory model comprises: Determine a third possibility that any two data points in the respiratory data set are adjacent upper and lower vertices of a cosine function corresponding to a Che-Kessler respiratory model; Determine a cosine function corresponding to the Che-Kessler breathing model based on the breathing depth of the two data points with the third greatest probability and the time of the two data points with the third greatest probability; A second possibility that all data points in the respiratory data set belong to the Che-Kerk respiratory model is determined based on a cosine function corresponding to the Che-Kerk respiratory model.

5. A heart rate detection method based on a wearable sensor according to claim 4, characterized in that: Determining that any two data points in the respiratory data set are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler respiratory model is a third possibility, including: Determine the preliminary possibility that the arbitrary two data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model based on the direction of the curvature radius of any two data points in the respiratory data set, the number of data points between the arbitrary two data points, the first possibility that the data points between the arbitrary two data points are initially noise points, the number of data points having the same value as the arbitrary two data points, and the first weight correction coefficient; Correcting the preliminary possibility using a second weight correction coefficient, thereby obtaining a third possibility that the arbitrary two data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model; Among them, the first weight correction coefficient is determined based on the direction of the curvature radius of the i-th data point between the any two data points; the second weight correction coefficient is determined based on the derivative value of the linear function of the time interval constructed by the any two data points.

6. A heart rate detection method based on a wearable sensor according to claim 5, characterized in that: The method comprises: If the curvature radius of the i-th data point between any two data points is in the direction If , the first weight correction coefficient is 1, otherwise, the first weight correction coefficient is 0; If the derivative value of the linear function of the time interval constructed by any two data points is less than 0, the second weight correction coefficient is 1; if the derivative value of the linear function of the time interval constructed by any two data points is greater than or equal to 0, the second weight correction coefficient is 0.

7. A heart rate detection method based on a wearable sensor according to claim 5, characterized in that: Determining the preliminary possibility that the two arbitrary data points are adjacent upper and lower vertices of the cosine function corresponding to the Che-Kessler breathing model includes: The preliminary probability is determined using the following formula: ； ； ； in, It indicates the preliminary possibility that any two data points OA and OB are adjacent upper and lower vertices of the cosine function corresponding to the Che-Schmidt breathing model. represents an exponential function with the natural constant e as the base, and max represents the maximum value. Indicates the direction of the curvature radius of the data point OA, Indicates the direction of the curvature radius of the data point OB, Indicates the number of data points with the same value as data point OA, Indicates the number of data points with the same value as data point OB. represents the first possibility that the jth data point between data points OA and OB is a noise point, m represents the number of data points between data points OA and OB, The first weight correction coefficient representing the first possibility that the jth data point between data points OA and OB is initially a noise point, It means that any two data points OA and OB are the corresponding upper and lower vertices of the cosine function under the reasonableness of the angle relationship. It indicates the reasonableness of the angle between any two data points OA and OB, which are the adjacent upper and lower vertices of the corresponding cosine function.

8. A heart rate detection method based on a wearable sensor according to claim 4, characterized in that: Determining a cosine function corresponding to the Che-Kessler breathing model based on the breathing depth of the two data points with the third greatest probability and the time of the two data points with the third greatest probability includes: Determine the breathing depth difference between the two data points with the third greatest probability as the amplitude of the cosine function, and determine the period of the cosine function using the time difference between the two data points with the third greatest probability; Constructing a reference formula using the amplitude of the cosine function, the period of the cosine function, the breathing depth of the two data points with the third greatest probability, and the time of the two data points with the third greatest probability; Based on the reference formula, a phase parameter and an offset parameter of a cosine function corresponding to the Che-Kessler breathing model are obtained; A cosine function corresponding to the Che-Kerk breathing model is constructed based on the phase parameter and the offset parameter.

9. A heart rate detection method based on a wearable sensor according to claim 4, characterized in that: Determining a second possibility that all data points in the respiratory data set belong to the Che-Kerk respiratory model based on the cosine function corresponding to the Che-Kerk respiratory model includes: Determining a fourth possibility that the current data point in the respiratory data set preliminarily belongs to the Che-Kessler respiratory model based on the cosine function corresponding to the Che-Kessler respiratory model, thereby determining a fourth possibility that all data points preliminarily belong to the Che-Kessler respiratory model; Determine the fourth possible correction coefficient based on the time corresponding to the current data point and the time corresponding to the current data point and the left and right adjacent data points; The second possibility is corrected based on the correction coefficient of the fourth possibility to obtain the second possibility that the current data point belongs to the Che-Kessler breathing model, thereby determining the second possibility that all data points belong to the Che-Kessler breathing model.

10. A heart rate detection method based on a wearable sensor according to claim 8, characterized in that: The cosine function corresponding to the Che-Kerk breathing model is constructed based on the phase parameter and the offset parameter, including: The cosine function corresponding to the Che-Schmidt breathing model is: ； in, Indicates the breathing depth of data point A The absolute value of the difference between the breathing depth at data point B and Indicates the time of data point A The time with data point B The absolute value of the time difference, T represents the time of any data point, M represents the breathing depth of the data point calculated by substituting the time of any data point into the cosine function corresponding to the Che-Kess breathing model, represents the phase parameter, Indicates the offset parameter.

Images