A short-term HRV measurement method based on machine learning

Abstract

The present invention provides a short-time HRV measurement method based on machine learning, which belongs to the field of machine learning technology, including: matching a time window function from a scene-function database according to a health assessment scenario of a target user, and collecting an electrocardiogram signal of the target user, and at the same time, capturing a user behavior image of the target user based on each acquisition moment after starting the time window function; obtaining acquisition signals based on all acquisition moments before the current moment, and analyzing the signal stability and signal representativeness of the acquisition signals; and dynamically adjusting the time window function based on the user behavior image at each acquisition moment before the current moment until the acquisition is completed, obtaining an initial HRV signal, and performing real-time evaluation and adjustment, obtaining a final HRV signal, and outputting it. The heart rate variability data collected in a shorter period of time can also replace the standard measurement of 5 minutes, and maintain a high evaluation accuracy and consistency.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a HRV short-term measurement method based on machine learning. Background Art

[0002] Heart rate variability (HRV) is an indicator that reflects the functional state of the autonomic nervous system and is widely used in health monitoring, stress assessment, and cardiovascular disease risk assessment. At present, the internationally recognized HRV assessment standard is based on 5 minutes of measurement data, which is regarded as the gold standard to ensure the stability and accuracy of the data. In traditional technical solutions, a 5-minute HRV measurement collects ECG signals, extracts the RR interval sequence, and then performs time domain, frequency domain or nonlinear analysis to obtain heart rate variability parameters. This standard assessment has been widely used in various health studies and applications, forming a mature method and assessment system.

[0003] Although 5-minute HRV measurement has been recognized as the gold standard, this measurement duration has obvious defects and inconveniences in many practical application scenarios. The main problems include:

[0004] Long measurement time: The 5-minute measurement time is too long for some application scenarios, such as emergency monitoring, sports health monitoring, and daily rapid health assessment. This results in the inability to provide effective information in a timely manner when rapid clinical feedback is urgently needed.

[0005] Poor user experience: For daily health monitoring users, 5 minutes of static measurement may not be convenient to perform, especially when evaluating on mobile devices. Users often cannot maintain a static state for a long time, which affects the accuracy of the measurement.

[0006] Application restrictions for specific groups: For the elderly, children or users with certain physical conditions, the 5-minute measurement process may be interrupted due to discomfort or intolerance, making it impossible to obtain accurate HRV data.

[0007] The 5-minute measurement is established as the gold standard for HRV, mainly because it can provide relatively stable and reliable data and eliminate instantaneous interference that may exist in short-term measurements. However, this also makes this method difficult to adapt to certain rapid response needs and mobile application scenarios. Since heart rate variability is essentially a complex reflection of autonomic nervous system activity, shorter measurements are often difficult to obtain reliable HRV indicators due to unstable signals or insufficient statistical samples. Therefore, how to ensure the accuracy and interchangeability of HRV parameters while shortening the measurement time has become a technical difficulty. In addition, HRV data measured for less than 5 minutes has not yet been clearly included in the standard scope in international guidelines, which makes research in this field lack recognized norms and references, and increases the technical difficulty of short-term measurement substitution.

[0008] Therefore, the present invention proposes a short-term HRV measurement method based on machine learning. Summary of the invention

[0009] The present invention provides a short-term HRV measurement method based on machine learning to solve the problems of the 5-minute standard measurement time being too long, poor user experience, and inability to adapt to the needs of rapid evaluation. The core of this technical route is to optimize the HRV calculation model, data collection and processing strategy, so that the heart rate variability data collected in a shorter period of time can also replace the 5-minute standard measurement and maintain high evaluation accuracy and consistency.

[0010] The present invention provides a method for short-term HRV measurement based on machine learning, comprising:

[0011] Step 1: According to the health assessment scenario of the target user, a time window function is matched from the scenario-function database, and the ECG signal of the target user is collected. At the same time, after the time window function is started, the user behavior image of the target user based on each collection moment is captured;

[0012] Step 2: Acquire the collected signals at all the collection moments before the current moment, and analyze the signal stability and signal representativeness of the collected signals;

[0013] Step 3: According to the signal stability, signal representativeness and the user behavior image at each acquisition moment before the current moment, the time window function is dynamically adjusted until the acquisition is completed to obtain an initial HRV signal;

[0014] Step 4: A signal quality classification algorithm based on machine learning is used to evaluate and adjust the initial HRV signal in real time to obtain a final HRV signal and output it.

[0015] Preferably, analyzing the signal stability and signal representativeness of the collected signal includes:

[0016] Comparing and analyzing the collected signals with representative signals of different typical types respectively;

[0017] Determine, based on the comparative analysis results, the representativeness of the collected signal for each typical type of signal;

[0018] At the same time, the collected signals are subjected to horizontal continuous analysis and vertical super-abnormal analysis to determine signal stability.

[0019] Preferably, dynamically adjusting the time window function includes:

[0020] Perform contour selection on the user behavior image at each acquisition moment to construct a contour sequence of the target user;

[0021] Obtaining a difference sequence based on the contour sequences at two consecutive acquisition moments, respectively, and inputting the difference sequence into a posture analysis model to determine a behavior change coefficient;

[0022] Obtaining a first influence coefficient matching the behavior change coefficient from a coefficient-influence comparison table;

[0023] Locking the prominent subsequences in the difference sequence, and determining a second influence coefficient based on the positional relationship between the body position of each prominent subsequence and the position of the device for collecting the electrocardiogram signal;

[0024] Determining a first number of collection expansion moments based on a first influence coefficient and a second influence coefficient at two different consecutive collection moments;

[0025] Optimizing the first quantity according to the signal stability and the signal representativeness to obtain a second quantity;

[0026] The time window function is dynamically adjusted according to the second quantity.

[0027] Preferably, determining the second influence coefficient includes:

[0028] A first position vector is constructed based on the body position of each prominent subsequence, and a relative position analysis is performed with a second position vector constructed based on the contour position of the body contour to obtain a comprehensive relative influence function on the device collecting the ECG signal. ,in, is the corresponding first position vector; is the corresponding second position vector;

[0029] Based on the comprehensive relative influence function , and combined with the position relationship, determine the second influence coefficient ;

[0030]

[0031] Wherein, ln represents the logarithmic function symbol; represents the absolute value of the i1th prominent subsequence; The correlation factor representing the position relationship between the position of the ith prominent subsequence and the position of the device for collecting ECG signals, with a value range of (0, 1); Indicates the total number of protruding subsequences present.

[0032] Preferably, determining the first number of acquisition expansion moments includes:

[0033] Obtaining a first fine-tuning factor according to a first ratio of the number of the protruding subsequences to the number of subsequences present in the contour sequence, and in combination with a second ratio of the sum of the absolute values ​​involved in the protruding subsequences to the sum of the absolute values ​​involved in the difference sequence;

[0034] According to the first fine-tuning factor, a proportion coefficient is respectively assigned to the first influence coefficient and the second influence coefficient, and a first quantity is obtained;

[0035]

[0036] in, is the corresponding first quantity; Represents a calculation function; Indicates that the calculation function N01 is rounded up; represents the first fine-tuning factor; Indicates the number of moments of the set maximum measurement time; Indicates the current time and the number of times before the current time; They represent moment analysis functions respectively; Represents the time instant number of the time window function itself; denote the first ratio and the second ratio respectively; Indicates the impact setting threshold; min and max represent the minimum and maximum values ​​respectively.

[0037] Preferably, dynamically adjusting the time window function according to the second quantity includes:

[0038] When the signal stability is greater than the preset stability, and one of all the signal representativenesses is greater than the preset representativeness of the corresponding typical type, the first quantity is regarded as the second quantity;

[0039] When the signal stability is greater than the preset stability, and there is no signal representativeness greater than the preset representativeness of the corresponding typical type among all signal representativeness, then The first quantity is adjusted to obtain a second quantity M02, wherein: Indicates the preset representativeness of the kth typical type; represents the signal representativeness of the kth typical type; m1 represents the number of typical types;

[0040] Otherwise, the first quantity is adjusted according to the difference between the preset stability and the signal stability to obtain a second quantity M02;

[0041] If M02+ , then based on Based on , the time window function is dynamically adjusted according to M02;

[0042] If M02+ , then based on Based on The time window function is dynamically adjusted.

[0043] Preferably, the signal quality classification algorithm based on machine learning performs real-time evaluation and adjustment on the initial HRV signal, including:

[0044] Performing noise filtering on the initial HRV signal;

[0045] Based on the adaptive filter, the filtered signal is optimized according to the pseudo error and the interference caused by the electrode to obtain a primary analysis signal;

[0046] A quality assessment is performed on each signal point in the primary analysis signal to obtain a final HRV signal.

[0047] Preferably, performing quality assessment on each signal point in the primary analysis signal to obtain a final HRV signal includes:

[0048] Divide each signal point into n01 consecutive time points based on the primary analysis signal to obtain an analysis segment including the corresponding signal point;

[0049] Performing jitter analysis and jamming analysis on the analysis segment to obtain a first analysis value;

[0050] Loading the primary analysis signal based on a standard waveform file to obtain an interference value for each signal point;

[0051] Based on the first capture time of the signal captured initially by the device collecting the ECG signal and the second capture time of the signal captured stably in history, and combined with the startup time of the corresponding device, an error correction function of the corresponding device is obtained;

[0052] According to the error correction function, and in combination with the position of the signal segment based on the primary analysis signal, an actual correction function is obtained;

[0053] According to the actual correction function and in combination with the interference value, the corresponding first analysis value is adjusted to obtain a second analysis value;

[0054] If the random consecutive n02 second analysis values ​​are all smaller than the preset analysis value, the corresponding signal point will be removed;

[0055] The curve is automatically redrawn according to the remaining signal points to obtain the final HRV signal.

[0056] Compared with the prior art, the beneficial effects of this application are as follows:

[0057] This is to solve the problems of the 5-minute standard measurement time being too long, poor user experience, and inability to adapt to the needs of rapid evaluation. The core of this technical route is to optimize the HRV calculation model, data collection and processing strategy, so that the heart rate variability data collected in a shorter period of time can also replace the 5-minute standard measurement and maintain a high evaluation accuracy and consistency.

[0058] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0059] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0061] Figure 1 The figure is a flow chart of a method for short-term HRV measurement based on machine learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0063] The present invention provides a method for short-term HRV measurement based on machine learning, such as Figure 1 As shown, including:

[0064] Step 1: According to the health assessment scenario of the target user, a time window function is matched from the scenario-function database, and the ECG signal of the target user is collected. At the same time, after the time window function is started, the user behavior image of the target user based on each collection moment is captured;

[0065] Step 2: Acquire the collected signals at all the collection moments before the current moment, and analyze the signal stability and signal representativeness of the collected signals;

[0066] Step 3: According to the signal stability, signal representativeness and the user behavior image at each acquisition moment before the current moment, the time window function is dynamically adjusted until the acquisition is completed to obtain an initial HRV signal;

[0067] Step 4: A signal quality classification algorithm based on machine learning is used to evaluate and adjust the initial HRV signal in real time to obtain a final HRV signal and output it.

[0068] In this embodiment, the health assessment scenario refers to scenarios such as sports rehabilitation, emergency warning, and daily health monitoring.

[0069] In this embodiment, the scenario-function database includes different health assessment scenarios and monitoring times set in advance based on the scenarios, wherein the time window function is: T=G (health assessment scenario).

[0070] In this embodiment, after the time window function is started, in order to determine the reliability of ECG signal monitoring, it is necessary to synchronously monitor the user's behavior, and the user's behavior is reflected based on the behavior image, and further realized through the posture change of the body contour.

[0071] In this embodiment, signal stability refers to whether the signal has a straight line or whether the amplitude is abnormal, that is, it does not conform to the common clinical signal changes, because signal interruption, sudden loss, etc. will reduce stability.

[0072] In this embodiment, the representativeness of the signal is obtained by comparing the signal with a typical type of signal, and the closer the match, the stronger its representativeness.

[0073] In this embodiment, the dynamic adjustment of the time window function refers to the adjustment of the time of the function, which can be shortening or lengthening the time, but the overall time must be less than 5 minutes, and is generally controlled within 1 to 3 minutes.

[0074] In this embodiment, the real-time evaluation and adjustment refers to removing signal points with poor quality in the signal and automatically redrawing the curve to obtain the final HRV signal.

[0075] The beneficial effects of the above technical solution are: to solve the problems of the 5-minute standard measurement time being too long, poor user experience, and inability to adapt to the needs of rapid evaluation. The core of this technical route is to optimize the HRV calculation model, data collection and processing strategy, so that the heart rate variability data collected in a shorter period of time can also replace the 5-minute standard measurement and maintain a high evaluation accuracy and consistency.

[0076] The present invention provides a short-term HRV measurement method based on machine learning, which analyzes the signal stability and signal representativeness of the collected signal, including:

[0077] Comparing and analyzing the collected signals with representative signals of different typical types respectively;

[0078] Determine, based on the comparative analysis results, the representativeness of the collected signal for each typical type of signal;

[0079] At the same time, the collected signals are subjected to horizontal continuous analysis and vertical super-abnormal analysis to determine signal stability.

[0080] In this embodiment, representative signals of the typical type include but are not limited to: normal electrocardiogram, sinus electrocardiogram, atrial premature beat electrocardiogram, atrial flutter electrocardiogram, atrial fibrillation electrocardiogram, ventricular premature beat electrocardiogram, etc.

[0081] In this embodiment, the signal representativeness refers to the similarity between the collected signal and a representative signal of a typical type. The more similar, the higher the representativeness, and the value range is 0 to 1.

[0082] In this embodiment, the horizontal continuity analysis refers to whether there is a signal interruption or loss, and the vertical super-abnormal analysis refers to whether the signal amplitude exceeds the set p times / minute.

[0083] Signal stability = lateral continuity coefficient × vertical super-abnormal coefficient = (number of lateral interruption and missing signal points / total signal points of collected signals) × (number of vertical super-abnormal signal points / total signal points of collected signals).

[0084] The beneficial effect of the above technical solution is: considering both representativeness and stability to provide a basis for the subsequent dynamic adjustment of the time function.

[0085] The present invention provides a method for short-term HRV measurement based on machine learning, which dynamically adjusts the time window function, including:

[0086] Perform contour selection on the user behavior image at each acquisition moment to construct a contour sequence of the target user;

[0087] Obtaining a difference sequence based on the contour sequences at two consecutive acquisition moments, respectively, and inputting the difference sequence into a posture analysis model to determine a behavior change coefficient;

[0088] Obtaining a first influence coefficient matching the behavior change coefficient from a coefficient-influence comparison table;

[0089] Locking the prominent subsequences in the difference sequence, and determining a second influence coefficient based on the positional relationship between the body position of each prominent subsequence and the position of the device for collecting the electrocardiogram signal;

[0090] Determining a first number of collection expansion moments based on a first influence coefficient and a second influence coefficient at two different consecutive collection moments;

[0091] Optimizing the first quantity according to the signal stability and the signal representativeness to obtain a second quantity;

[0092] The time window function is dynamically adjusted according to the second quantity.

[0093] In this embodiment, the contour selection refers to the edge lines of the screened user's body contour, and the contour sequence = {edge position value of each body point}.

[0094] In this embodiment, for example, there are time t1 and time t2, and the difference sequence is the difference between the edge position value based on each body point at time t1 and time t2.

[0095] In this embodiment, the posture analysis model is obtained by training the neural network model based on the behavioral differences of actions with different changes and the behavioral change coefficient of the device's signal collection caused by the behavioral differences. Therefore, the behavior change coefficient can be directly obtained, and the value range is 0 to 1.

[0096] In this embodiment, the coefficient-influence comparison table includes different behavior change coefficients and corresponding influence coefficients, which are preset.

[0097] In this embodiment, a prominent subsequence refers to a subsequence whose absolute value of the difference exceeds the preset value of the corresponding body point. For example, the shoulder point A1 is raised by 10 cm. At this time, the preset value is 8 cm, which is regarded as a prominent subsequence.

[0098] The beneficial effect of the above technical solution is: based on the combination of the neural network model and the impact analysis of the prominent subsequence, the number of moments that need to be expanded is comprehensively determined, and the number is optimized in combination with signal stability and signal representativeness to ensure the rationality of the expansion, and then determine the reliability of the function adjustment, so that the collected ECG signal has analysis value.

[0099] The present invention provides a short-term HRV measurement method based on machine learning, which determines a second influence coefficient, including:

[0100] A first position vector is constructed based on the body position of each prominent subsequence, and a relative position analysis is performed with a second position vector constructed based on the contour position of the body contour to obtain a comprehensive relative influence function on the device collecting the ECG signal. ,in, is the corresponding first position vector; is the corresponding second position vector;

[0101] Based on the comprehensive relative influence function , and combined with the position relationship, determine the second influence coefficient ;

[0102]

[0103] Wherein, ln represents the logarithmic function symbol; represents the absolute value of the i1th prominent subsequence; The correlation factor representing the position relationship between the position of the ith prominent subsequence and the position of the device for collecting ECG signals, with a value range of (0, 1); Indicates the total number of protruding subsequences present.

[0104] In this embodiment, the first position vector = {the body position of each protrusion in the sequence}.

[0105] In this embodiment, the second position vector={all contour positions of the body contour}.

[0106] Among them, the relative position analysis is to analyze the influence of the existence of the first position vector on the second position vector, and the first position vector and the second position vector can be input into the vector analysis model to obtain them, wherein the vector analysis model is based on the relative position relationship of different vectors and the relative influence function under the relative position relationship for sample training.

[0107] The beneficial effect of the above technical solution is: by highlighting the position vector of the subsequence and the position vector of the contour position, the relative influence function is determined, and then the second influence coefficient is obtained in combination with the position relationship, providing an analysis basis for dynamically adjusting the time.

[0108] The present invention provides a method for short-term HRV measurement based on machine learning, which determines a first number of acquisition expansion moments, including:

[0109] Obtaining a first fine-tuning factor according to a first ratio of the number of the protruding subsequences to the number of subsequences present in the contour sequence, and in combination with a second ratio of the sum of the absolute values ​​involved in the protruding subsequences to the sum of the absolute values ​​involved in the difference sequence;

[0110] According to the first fine-tuning factor, a proportion coefficient is respectively assigned to the first influence coefficient and the second influence coefficient, and a first quantity is obtained;

[0111]

[0112] in, is the corresponding first quantity; Represents a calculation function; Indicates that the calculation function N01 is rounded up; represents the first fine-tuning factor; Indicates the number of moments of the set maximum measurement time; Indicates the current time and the number of times before the current time; They represent moment analysis functions respectively; Represents the time instant number of the time window function itself; denote the first ratio and the second ratio respectively; Indicates the impact setting threshold; min and max represent the minimum and maximum values ​​respectively.

[0113] In this embodiment, the first fine-tuning factor=the result of multiplying the first ratio by the second ratio and calculating the mean square.

[0114] In this embodiment, the proportion coefficient is .

[0115] The beneficial effect of the above technical solution is: the ratio is determined from the number of prominent subsequences and the absolute value of the difference, and then the fine-tuning factor is obtained, and then combined with the two influence coefficients to obtain the first number, which provides a basis for moment adjustment again.

[0116] The present invention provides a method for short-term HRV measurement based on machine learning, which dynamically adjusts the time window function according to the second quantity, including:

[0117] When the signal stability is greater than the preset stability, and one of all the signal representativenesses is greater than the preset representativeness of the corresponding typical type, the first quantity is regarded as the second quantity;

[0118] When the signal stability is greater than the preset stability, and there is no signal representativeness greater than the preset representativeness of the corresponding typical type among all signal representativeness, then The first quantity is adjusted to obtain a second quantity M02, wherein: Indicates the preset representativeness of the kth typical type; represents the signal representativeness of the kth typical type; m1 represents the number of typical types;

[0119] Otherwise, the first quantity is adjusted according to the difference between the preset stability and the signal stability to obtain a second quantity M02;

[0120] If M02+ , then based on Based on , the time window function is dynamically adjusted according to M02;

[0121] If M02+ , then based on Based on The time window function is dynamically adjusted.

[0122] In this embodiment, the preset representative value is 0.6.

[0123] In this embodiment, M02=ceiling(M01×(1+variance of the differences in the stability of all signals)).

[0124] In this embodiment, the number of acquisition time moments corresponding to the original time window function is At this time, when M02+ , continue to collect signals for M02 moments after the current moment.

[0125] When M02+ When the current time is reached, the collection continues A signal of the moment.

[0126] The beneficial effect of the above technical solution is: the first quantity is judged in combination with signal stability and signal representativeness to achieve dynamic adjustment of the time window function and ensure the reliability of signal acquisition.

[0127] The present invention provides a short-term HRV measurement method based on machine learning, which performs real-time evaluation and adjustment on the initial HRV signal based on a signal quality classification algorithm based on machine learning, including:

[0128] Performing noise filtering on the initial HRV signal;

[0129] Based on the adaptive filter, the filtered signal is optimized according to the pseudo error and the interference caused by the electrode to obtain a primary analysis signal;

[0130] A quality assessment is performed on each signal point in the primary analysis signal to obtain a final HRV signal.

[0131] In this embodiment, noise filtering belongs to the prior art.

[0132] The beneficial effect of the above technical solution is that in short-term data collection, the heart rate signal is more susceptible to external interference, and then by adding noise filtering, an adaptive filter is used to remove artifacts and electrode interference, ensuring that the final calculated HRV data is sufficiently accurate and reliable. A signal quality classification algorithm based on machine learning is used to evaluate the quality of the ECG signal in real time, and low-quality collected data is eliminated to ensure that the final output HRV evaluation result is accurate and reliable.

[0133] The present invention provides a method for short-term HRV measurement based on machine learning, which performs quality evaluation on each signal point in the primary analysis signal to obtain a final HRV signal, including:

[0134] Divide each signal point into n01 consecutive time points based on the primary analysis signal to obtain an analysis segment including the corresponding signal point;

[0135] Performing jitter analysis and jamming analysis on the analysis segment to obtain a first analysis value;

[0136] Loading the primary analysis signal based on a standard waveform file to obtain an interference value for each signal point;

[0137] Based on the first capture time of the signal captured initially by the device collecting the ECG signal and the second capture time of the signal captured stably in history, and combined with the startup time of the corresponding device, an error correction function of the corresponding device is obtained;

[0138] According to the error correction function, and in combination with the position of the signal segment based on the primary analysis signal, an actual correction function is obtained;

[0139] According to the actual correction function and in combination with the interference value, the corresponding first analysis value is adjusted to obtain a second analysis value;

[0140] If the random consecutive n02 second analysis values ​​are all smaller than the preset analysis value, the corresponding signal point will be removed;

[0141] The curve is automatically redrawn according to the remaining signal points to obtain the final HRV signal.

[0142] In this embodiment, the continuous n01 time points refer to a random signal segment including a signal point, which may be the first point, the last point, the middle point, etc. with the signal point as the first point, the last point, the middle point, etc., and it is sufficient that there are continuous n01 time points.

[0143] In this embodiment, the analysis segment includes n01 consecutive time points including the corresponding signal points.

[0144] In this embodiment, jitter analysis refers to the jitter of the signal, which can be captured using an oscilloscope, and jam analysis refers to the situation where the signal is in a continuous straight line, that is, there is no fluctuation.

[0145] In this embodiment, the first analysis value=the sum of the jitter amplitudes of the analysis segment / the sum of the total amplitudes of the analysis segment+the number of jamming signal points / n01.

[0146] In this embodiment, the standard waveform file is a file that is pre-loaded for interference analysis of the signal, so the interference value existing at each signal point can be obtained by directly loading the application.

[0147] In this embodiment, since the operation of the device is not yet stable after the startup is set, the time RT0 when the device starts, the time RT1 when the device starts to capture the signal, and the time RT2 when the device captures the stable signal are obtained, wherein the startup time is before the time when the signal starts to be captured, and the time when the signal starts to be captured is before the time when the stable signal is captured.

[0148] In this embodiment, the error correction function = WX (RT0, RT1, RT2).

[0149] In this embodiment, the actual correction function = Y (error correction function, the signal segment is based on the position of a single analysis signal), and the value of the actual correction function is matched from a function-value comparison table, which contains different actual correction functions and corresponding values ​​based on RT0, RT1, and RT2, and the position of the signal segment.

[0150] In this embodiment, the second analysis value=the first analysis value×(1+actual correction function+interference value / maximum interference setting value).

[0151] In this embodiment, the preset analysis value is set in advance.

[0152] In this embodiment, the consecutive n02 refers to the second analysis values ​​of consecutive n02 signal points.

[0153] In this embodiment, automatic curve drawing is a function that can be automatically implemented by the software.

[0154] The beneficial effect of the above technical solution is: by continuously dividing n01 time points to perform jitter and jamming analysis, a calculation basis is provided, and then the error correction function constructed at the capture moment, the interference value and the position of the signal segment are combined to comprehensively adjust the first analysis value to ensure the reliability of the final HRV signal.

[0155] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A short-term HRV measurement method based on machine learning, characterized in that: include: Step 1: According to the health assessment scenario of the target user, a time window function is matched from the scenario-function database, and the ECG signal of the target user is collected. At the same time, after the time window function is started, the user behavior image of the target user based on each collection moment is captured; Step 2: Acquire the collected signals at all the collection moments before the current moment, and analyze the signal stability and signal representativeness of the collected signals; Step 3: According to the signal stability, signal representativeness and the user behavior image at each acquisition moment before the current moment, the time window function is dynamically adjusted until the acquisition is completed to obtain an initial HRV signal; Step 4: A signal quality classification algorithm based on machine learning is used to evaluate and adjust the initial HRV signal in real time to obtain a final HRV signal and output it.

2. The method for short-term HRV measurement based on machine learning according to claim 1, characterized in that: Analyze the signal stability and signal representativeness of the collected signal, including: Comparing and analyzing the collected signals with representative signals of different typical types respectively; Determine, based on the comparative analysis results, the representativeness of the collected signal for each typical type of signal; At the same time, the collected signals are subjected to horizontal continuous analysis and vertical super-abnormal analysis to determine signal stability.

3. The method for short-term HRV measurement based on machine learning according to claim 1, characterized in that: Dynamically adjusting the time window function includes: Perform contour selection on the user behavior image at each acquisition moment to construct a contour sequence of the target user; Obtaining a difference sequence based on the contour sequences at two consecutive acquisition moments, respectively, and inputting the difference sequence into a posture analysis model to determine a behavior change coefficient; Obtaining a first influence coefficient matching the behavior change coefficient from a coefficient-influence comparison table; Locking the prominent subsequences in the difference sequence, and determining a second influence coefficient depending on the positional relationship between the body position of each prominent subsequence and the position of the device for collecting the electrocardiogram signal; Determining a first number of collection expansion moments based on a first influence coefficient and a second influence coefficient at two different consecutive collection moments; Optimizing the first quantity according to the signal stability and the signal representativeness to obtain a second quantity; The time window function is dynamically adjusted according to the second quantity.

4. The method for short-term HRV measurement based on machine learning according to claim 3, characterized in that: Determine the second influence coefficient, including: A first position vector is constructed based on the body position of each prominent subsequence, and a relative position analysis is performed with a second position vector constructed based on the contour position of the body contour to obtain a comprehensive relative influence function on the device collecting the ECG signal. ,in, is the corresponding first position vector; is the corresponding second position vector; Based on the comprehensive relative influence function , and combined with the position relationship, determine the second influence coefficient ; ； Wherein, ln represents the logarithmic function symbol; represents the absolute value of the i1th prominent subsequence; The correlation factor representing the position relationship between the position of the ith prominent subsequence and the position of the device for collecting ECG signals, with a value range of (0, 1); Indicates the total number of protruding subsequences present.

5. The method for short-term HRV measurement based on machine learning according to claim 3, characterized in that: Determining a first number of acquisition expansion moments includes: Obtaining a first fine-tuning factor according to a first ratio of the number of the protruding subsequences to the number of subsequences present in the contour sequence, and in combination with a second ratio of the sum of the absolute values ​​involved in the protruding subsequences to the sum of the absolute values ​​involved in the difference sequence; According to the first fine-tuning factor, a proportion coefficient is respectively assigned to the first influence coefficient and the second influence coefficient, and a first quantity is obtained; ； in, is the corresponding first quantity; Represents a calculation function; Indicates that the calculation function N01 is rounded up; represents the first fine-tuning factor; Indicates the number of moments of the set maximum measurement time; Indicates the current time and the number of times before the current time; They represent moment analysis functions respectively; Represents the time instant number of the time window function itself; denote the first ratio and the second ratio respectively; Indicates the impact setting threshold; min and max represent the minimum and maximum values ​​respectively.

6. The method for short-term HRV measurement based on machine learning according to claim 3, characterized in that: Dynamically adjusting the time window function according to the second quantity includes: When the signal stability is greater than the preset stability, and one of all the signal representativenesses is greater than the preset representativeness of the corresponding typical type, the first quantity is regarded as the second quantity; When the signal stability is greater than the preset stability, and there is no signal representativeness greater than the preset representativeness of the corresponding typical type among all signal representativeness, then The first quantity is adjusted to obtain a second quantity M02, wherein: Indicates the preset representativeness of the kth typical type; represents the signal representativeness of the kth typical type; m1 represents the number of typical types; Otherwise, the first quantity is adjusted according to the difference between the preset stability and the signal stability to obtain a second quantity M02; If M02+ , then based on Based on , the time window function is dynamically adjusted according to M02; If M02+ , then based on Based on The time window function is dynamically adjusted.

7. The method for short-term HRV measurement based on machine learning according to claim 1, characterized in that: The signal quality classification algorithm based on machine learning performs real-time evaluation and adjustment on the initial HRV signal, including: Performing noise filtering on the initial HRV signal; Based on the adaptive filter, the filtered signal is optimized according to the pseudo error and the interference caused by the electrode to obtain a primary analysis signal; A quality assessment is performed on each signal point in the primary analysis signal to obtain a final HRV signal.

8. The method for short-term HRV measurement based on machine learning according to claim 7, characterized in that: Performing quality assessment on each signal point in the primary analysis signal to obtain a final HRV signal includes: Divide each signal point into n01 consecutive time points based on the primary analysis signal to obtain an analysis segment including the corresponding signal point; Performing jitter analysis and jamming analysis on the analysis segment to obtain a first analysis value; Loading the primary analysis signal based on a standard waveform file to obtain an interference value for each signal point; Based on the first capture time of the signal captured initially by the device collecting the ECG signal and the second capture time of the signal captured stably in history, and combined with the startup time of the corresponding device, an error correction function of the corresponding device is obtained; According to the error correction function, and in combination with the position of the signal segment based on the primary analysis signal, an actual correction function is obtained; According to the actual correction function and in combination with the interference value, the corresponding first analysis value is adjusted to obtain a second analysis value; If the random consecutive n02 second analysis values ​​are all smaller than the preset analysis value, the corresponding signal point will be removed; The curve is automatically redrawn according to the remaining signal points to obtain the final HRV signal.

Images