Health monitoring analysis early warning method and system based on multi-modal data fusion

Abstract

The invention relates to the technical field of medical health monitoring, and discloses a multi-modal data fusion health monitoring analysis early warning method and system, and the method comprises the steps: synchronously collecting motion and heart rate data streams through a wearable device, generating a coupling vector of the motion intensity and the heart rate change rate in a short time window, and carrying out the long-time accumulation to form a coupling vector cluster, and judging the physiological regulation stability by evaluating the change of the morphological dispersion. Micro-motion is converted into a natural probe, through dynamic evolution of a motion-heart rate coupling relation, transition from observation of vital sign values to evaluation of autonomic nerve regulation ability is achieved, early decline of body functions can be captured before conventional indexes are abnormal, and meanwhile evaluation reliability in a complex environment is ensured through frequency domain feature filtering.

Description

Technical Field

[0001] The present invention relates to a health monitoring, analysis and early warning method and system for multimodal data fusion, belonging to the technical field of medical health monitoring. Background Art

[0002] Existing technologies usually achieve abnormal alarms by continuously collecting physiological parameters such as heart rate and blood oxygen, and comparing them with preset standard ranges. However, this type of method has fundamental limitations: it is essentially a passive observation of physiological regulation results, rather than an assessment of the body's intrinsic ability to maintain homeostasis; when elderly people living alone are in the incubation period of the disease, the autonomic nervous system has begun to bear the load and the regulatory ability has undergone a hidden decline, but the conventional vital signs still show normal values, resulting in the system being unable to capture early risks.

[0003] Specifically, existing technologies have three main deficiencies: 1. Excessive focus on the absolute values ​​of vital signs and ignoring the dynamic stability of the regulation process, resulting in a lag in the key warning window; 2. Non-physiological vibrations in complex scenarios, such as vehicle bumps, will contaminate motion signals and cause misjudgment; 3. Simply quantifying discreteness cannot distinguish between hyperactive and inhibitory instability modes, weakening the value of intervention guidance.

[0004] The industry has attempted to improve accuracy by adding sensors or strengthening algorithms, but this has exacerbated the contradiction between power consumption and cost and failed to break through the essential limitations of the observation method. With the surge in demand for home health management, how to achieve early assessment of regulatory capabilities through low-power wearable devices has become a technical problem to be solved by this invention. Summary of the Invention

[0005] The present invention provides a health monitoring, analysis and early warning method and system for multimodal data fusion, the main purpose of which is to solve the problem that the existing technology cannot capture the early decline of the body's regulatory ability through low-power solutions.

[0006] To achieve the above objectives, the present invention provides a multimodal data fusion health monitoring, analysis and early warning method comprising the following steps: Step a: synchronously collecting a first physiological data stream and a second physiological data stream through a wearable device, wherein the first physiological data stream includes user motion information and the second physiological data stream includes user heart rate information, and the collection is performed in a low-frequency non-uniform manner; Step b: performing frequency domain analysis on the first physiological data stream within a specific short time window to determine whether the frequency distribution of its motion energy conforms to the physiological low-frequency characteristics; Step c: determining an exercise intensity scalar based on the first physiological data stream and a heart rate variability scalar based on the second physiological data stream only when the frequency distribution of the exercise energy conforms to the physiological low-frequency characteristics; Step d, pairing the exercise intensity scalar and the heart rate variability scalar to form a coupling vector that characterizes the physiological coupling response within the short time window; Step e: accumulating a series of coupling vectors over a specific long period of time to form a coupling vector cluster, where the coupling vector cluster reflects the intrinsic pattern of the user's physiological activity and heart rate response; Step f, evaluating the morphological discreteness of the coupling vector bundle, and judging the stability of the user's physiological coupling regulation based on the changing trend of the morphological discreteness in continuous time periods; Step g: When the physiological coupling regulation stability shows a decline trend, health risk warning information is generated.

[0007] Preferably, in step a, the first physiological data stream is acquired by a three-axis accelerometer sensor, and the second physiological data stream is acquired by a photoplethysmography sensor.

[0008] Preferably, in step c, the exercise intensity scalar is the variance of the acceleration signal in a specific short time window, and the heart rate variability scalar is the slope of the heart rate sequence in the specific short time window.

[0009] Preferably, in step b, the frequency domain analysis is specifically to perform a short-time Fourier transform on the first physiological data stream, and to determine whether the energy proportion of the high-frequency part in the short-time Fourier transform result is lower than a specific judgment threshold; in step f, the method for evaluating the morphological discreteness of the coupled vector bundle is specifically to calculate the convex hull area of ​​the coupled vector bundle in a two-dimensional coordinate system.

[0010] Preferably, the method further comprises a personal baseline self-calibration step: continuously recording the convex hull area and calculating its statistical average value over an ultra-long period of twenty-four to seventy-two hours, and using the statistical average value as a personal benchmark for judging the stability of physiological coupling regulation.

[0011] Preferably, in step g, the generation of the warning information is triggered when the convex hull area of ​​the current time window is higher than the alarm threshold determined by the personal benchmark for at least one hour.

[0012] Preferably, in the two-dimensional coordinate system of the coupled vector bundle, the X-axis represents exercise intensity and the Y-axis represents heart rate variability, which is divided into the following functional quadrants: a high-response area, in which the heart rate variability is higher than fifty percent of its average value in the resting state; a sluggish or disjointed area, in which the absolute value of the heart rate variability is less than five percent of its average value in the resting state.

[0013] Preferably, in step g, when it is determined that the physiological coupling regulation stability shows a declining trend, the step further includes: counting the number of coupling vectors falling into the high response area and the number of coupling vectors that fall into the sluggish or disjointed regions By comparison and The relative size of the physiological coupling regulation instability is used to determine the pattern characteristics. The judgment of the pattern characteristics is based on the following criteria: , it is determined to be hyperactive instability; if , it is determined to be an inhibitory instability, where is a preset ratio threshold greater than 1.

[0014] Preferably, the collection is performed in a low-frequency non-uniform manner, specifically by randomly waking up the device for five to ten seconds every minute to collect data.

[0015] A multimodal data fusion health monitoring, analysis and early warning system, including: A data acquisition module configured to synchronously acquire a first physiological data stream of user motion information and a second physiological data stream of user heart rate information, wherein the acquisition is performed in a low-frequency non-uniform manner; a frequency analysis module configured to perform frequency domain analysis on the first physiological data stream within a specific short time window to determine whether the frequency distribution of its motion energy conforms to the physiological low-frequency characteristics; a coupling vector generation module configured to determine an exercise intensity scalar based on the first physiological data stream and a heart rate variability scalar based on the second physiological data stream only when the frequency distribution of the exercise energy conforms to the physiological low-frequency characteristics, and to pair the exercise intensity scalar and the heart rate variability scalar to form a coupling vector representing the physiological coupling response within the short time window; A coupling vector cluster building module is configured to accumulate a series of coupling vectors over a specific long period of time to form a coupling vector cluster, where the coupling vector cluster reflects an intrinsic pattern of the user's physiological activity and heart rate response; a stability evaluation module configured to evaluate the morphological discreteness of the coupling vector bundle and determine the stability of the user's physiological coupling regulation based on a changing trend of the morphological discreteness within a continuous time period; The early warning generation module is configured to generate health risk early warning information when the physiological coupling regulation stability shows a decline trend.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. By converting micro-movements into natural probes, the system no longer relies on static values ​​of vital signs, but instead captures the coupling relationship between the exercise intensity scalar and the heart rate variability scalar. When these coupled vectors form a vector bundle over a long period of time, the evolution of their morphological discreteness directly reflects the regulatory stability of the autonomic nervous system. This transformation from observation results to assessment capabilities enables health monitoring to penetrate superficial data and capture the early decline of body function before conventional physiological indicators become abnormal.

[0017] 2. Traditional alarms rely on threshold breaches, while this invention, based on a dynamic assessment of the convex hull area trend, anchors the warning logic to the instability of an individual's own regulatory mode. When the vector bundle shifts from a compact form to a diffuse distribution, the system identifies a systematic decline in regulatory capacity. This mechanism, triggered by a loss of stability, avoids sensitive reliance on absolute value fluctuations, shifting the warning window from the onset of symptoms to the initial stage of dysfunction, creating a golden opportunity for intervention in chronic disease management. The newly added frequency-domain fingerprint identification mechanism forms a synergistic closed loop with this invention. By performing physiological low-frequency feature analysis on the acceleration signal, the system filters out non-physiological vibration interference before generating the coupling vector. This source filtering based on frequency-domain energy distribution eliminates hardware costs while ensuring the purity of the exercise-heart rate coupling relationship, ensuring the credibility of the stability assessment conclusions in complex scenarios such as buses and near electrical appliances.

[0018] 3. When the convex hull area expands abnormally, the system further activates the quadrant density statistics mechanism. By statistically analyzing the distribution density differences of vector bundles in high-response and slow-response areas, the system decomposes the single unstable signal into qualitative characteristics of hyperactivity or inhibition. This mechanism reuses existing data structures and achieves initial screening of pathological tendencies through partition counting alone, providing actionable guidance for family health management. The core mechanism requires only a basic combination of accelerometers and PPG sensors, replacing complex models with underlying operations such as variance, slope, and convex hull area. Frequency domain analysis uses an 8-point short-time FFT for interference filtering, and partition statistics require only counter accumulation, achieving ultimate utilization of universal hardware and achieving low-energy, high-quality early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a timing diagram of the multimodal data fusion health monitoring, analysis and early warning method of the present invention; Figure 2 This is a graph showing the relationship between exercise intensity and heart rate change rate of the present invention; Figure 3 This is a flow chart of the multimodal data fusion health monitoring and early warning system of the present invention.

[0020] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0022] This embodiment of the present application provides a health monitoring, analysis and early warning method for multimodal data fusion, comprising the following steps: Step a: synchronously collecting a first physiological data stream and a second physiological data stream through a wearable device, wherein the first physiological data stream includes user motion information and the second physiological data stream includes user heart rate information, and the collection is performed in a low-frequency non-uniform manner; Step b: performing frequency domain analysis on the first physiological data stream within a specific short time window to determine whether the frequency distribution of its motion energy conforms to the physiological low-frequency characteristics; Step c: determining an exercise intensity scalar based on the first physiological data stream and a heart rate variability scalar based on the second physiological data stream only when the frequency distribution of the exercise energy conforms to the physiological low-frequency characteristics; Step d, pairing the exercise intensity scalar and the heart rate variability scalar to form a coupling vector that characterizes the physiological coupling response within the short time window; Step e: accumulating a series of coupling vectors over a specific long period of time to form a coupling vector cluster, where the coupling vector cluster reflects the intrinsic pattern of the user's physiological activity and heart rate response; Step f, evaluating the morphological discreteness of the coupling vector bundle, and judging the stability of the user's physiological coupling regulation based on the changing trend of the morphological discreteness in continuous time periods; Step g: When the physiological coupling regulation stability shows a decline trend, health risk warning information is generated.

[0023] Preferably, in step a, the first physiological data stream is acquired by a three-axis accelerometer sensor, and the second physiological data stream is acquired by a photoplethysmography sensor.

[0024] Preferably, in step c, the exercise intensity scalar is the variance of the acceleration signal in a specific short time window, and the heart rate variability scalar is the slope of the heart rate sequence in the specific short time window.

[0025] Preferably, in step b, the frequency domain analysis is specifically to perform a short-time Fourier transform on the first physiological data stream, and to determine whether the energy proportion of the high-frequency part in the short-time Fourier transform result is lower than a specific judgment threshold; in step f, the method for evaluating the morphological discreteness of the coupled vector bundle is specifically to calculate the convex hull area of ​​the coupled vector bundle in a two-dimensional coordinate system.

[0026] Preferably, the method further includes a personal baseline self-calibration step: continuously recording the convex hull area and calculating its statistical average over an ultra-long period of twenty-four to seventy-two hours, and using the statistical average as a personal benchmark for judging the stability of physiological coupling regulation.

[0027] Preferably, in step g, the generation of the warning information is triggered when the convex hull area of ​​the current time window is higher than the alarm threshold determined by the personal benchmark for at least one hour.

[0028] Preferably, in the two-dimensional coordinate system of the coupled vector bundle, the X-axis represents the exercise intensity and the Y-axis represents the heart rate variability, which is divided into the following functional quadrants: a high response area, in which the heart rate variability is higher than 50% of its average value in the resting state; a sluggish or disjointed area, in which the absolute value of the heart rate variability is less than 5% of its average value in the resting state. In the present invention, the convex hull area is proposed as an indicator to characterize the degree of discreteness of the coupled vector bundle, mainly based on the following physiological regulation principle: when an individual is in a healthy state, there is a relatively consistent and regular synergistic relationship between their exercise behavior and heart rate response, which presents a relatively compact distribution form in the coupled vector coordinate system; and when When the autonomic nervous system's regulatory capacity declines, the body's heart rate response to the same type of exercise stimulation becomes unstable, delayed, or has abnormal amplitude, resulting in an increased degree of spatial divergence of the coupling vector. The convex hull area, as a spatial measure of the minimum envelope area, can directly reflect the overall distribution trend and degree of convergence of the coupling response. By continuously recording and comparing the changing trajectory of this area, the evolution of individual regulatory stability can be dynamically perceived without relying on absolute numerical fluctuations. This method has good physiological rationality and engineering operability, and also helps to complete early risk assessment without the need for complex model reasoning. All of these are extended implementation methods known to ordinary technicians in this field.

[0029] Preferably, in step g, when it is determined that the physiological coupling regulation stability shows a declining trend, the step further includes: counting the number of coupling vectors falling into the high response area and the number of coupling vectors that fall into the sluggish or disjointed regions By comparison and The relative size of the physiological coupling regulation instability is used to determine the pattern characteristics. The judgment of the pattern characteristics is based on the following criteria: , it is determined to be hyperactive instability; if , it is determined to be an inhibitory instability, where is a preset ratio threshold greater than 1.

[0030] Preferably, the collection is performed in a low-frequency non-uniform manner, specifically by randomly waking up the device for five to ten seconds every minute to collect data.

[0031] A multimodal data fusion health monitoring, analysis and early warning system includes: a data acquisition module configured to synchronously acquire a first physiological data stream of user motion information and a second physiological data stream of user heart rate information, the acquisition being performed in a low-frequency non-uniform manner; a frequency analysis module configured to perform frequency domain analysis on the first physiological data stream within a specific short time window to determine whether the frequency distribution of its motion energy conforms to the physiological low-frequency characteristics; a coupling vector generation module configured to determine an exercise intensity scalar based on the first physiological data stream and a heart rate variability scalar based on the second physiological data stream only when the frequency distribution of the motion energy conforms to the physiological low-frequency characteristics, and to pair the exercise intensity scalar and the heart rate variability scalar to form a coupling vector representing the physiological coupling response within the short time window. ; The coupling vector cluster construction module is configured to accumulate a series of coupling vectors over a specific long period of time to form a coupling vector cluster, which reflects the intrinsic pattern of the user's physiological activities and heart rate responses; the stability evaluation module is configured to evaluate the morphological discreteness of the coupling vector cluster, and judge the user's physiological coupling regulation stability based on the changing trend of the morphological discreteness in a continuous time period; the early warning generation module is configured to generate health risk early warning information when the physiological coupling regulation stability shows a declining trend; at the same time, the heart rate change rate refers to the heart rate change trend calculated from the continuous pulse wave intervals within a specific time window. The calculation is based on the heart rate sampling result once per second, and the time granularity is accurate to the second level. This is used as a reference for the changing trend, which is conducive to capturing rapid regulation responses. Furthermore, paired analysis of exercise intensity and heart rate variability is performed based on unified time segments, ensuring comparability and consistency of physiological coupling characteristics across the timeline. During long-term monitoring, an hourly updated evaluation mechanism is employed, combined with statistical methods to extract patterns in the coupling vector bundle. For example, the discrete trends formed by an individual's long-term data can be dynamically compared with the current state to avoid interference from static indicators. In specific implementations, the system can set reference ranges based on actual test samples. For example, in certain scenarios, when the spatial discreteness of the coupling vector bundle exceeds a specified percentage of the individual's historical mean for multiple consecutive time periods, this is considered a signal of decreased stability. The system thus triggers an early warning and identifies abnormal patterns in the coupling response, providing clear judgment basis and actionable adjustment recommendations for subsequent health management. Furthermore, the system generates evaluation results that not only reflect the current state but also focus on the dynamic evolution of the regulatory process. In particular, even when resting physiological data remain normal, subtle fluctuations in the consistency between exercise and heart rate responses can be used to early detect potential regulatory abnormalities. These are all extended implementations known to those skilled in the art.

[0032] Example 1: This example proposes a health monitoring, analysis and early warning method and system for multimodal data fusion, which aims to evaluate the early changes in the body's regulatory ability with the help of low-power wearable devices without relying on the absolute values ​​of traditional vital signs, and generate health risk early warning information based on this. First, the wearable device synchronously collects the user's motion data stream and heart rate data stream. Specifically, the first physiological data stream is collected by a three-axis accelerometer sensor to reflect the user's micro-motion information; the second physiological data stream is obtained by a photoelectric volume pulse wave sensor to represent the user's heart rate dynamics. The data collection process adopts low-frequency non- The uniform wake-up strategy means that the device randomly starts the data acquisition module every minute and runs it continuously for 5 to 10 seconds. This strategy effectively reduces energy consumption while also meeting the time resolution and data coverage required for health monitoring. The collected motion data is sent to the frequency domain analysis process within a specific short time window; the present invention uses short-time Fourier transform to process the first physiological data stream to determine whether the frequency distribution of its motion energy has physiological low-frequency characteristics. The physiological low-frequency characteristics referred to here refer to the proportion of high-frequency energy in the spectrum being lower than the set judgment threshold. Only when this condition is met will the subsequent processing steps be continued.

[0033] After passing the above-mentioned frequency domain screening, the present invention calculates two scalar indicators representing physiological responses based on acceleration data and heart rate data respectively: one is the exercise intensity scalar, specifically the variance of the acceleration signal in the short time window; the other is the heart rate change rate scalar, which is the slope of the heart rate sequence in the same time window. After pairing, the above two scalars constitute a coupling vector that characterizes the physiological coupling response state in the short time window; within a set long time period, the system continuously accumulates the above-mentioned coupling vectors, and then forms a coupling vector cluster. The coupling vector cluster is a collection of motion and heart rate coupling characteristics in multiple time segments, reflecting the inherent regularity between the user's physiological activities and autonomic nervous responses, and forming the basis for subsequent stability evaluation; in order to evaluate the user's current physiological regulation stability, the present invention calculates the coupling vector. The convex hull area formed by the resultant vector bundle in a two-dimensional coordinate system is used to characterize its morphological discreteness. In this coordinate system, the X-axis represents exercise intensity and the Y-axis represents heart rate variability. As the stability of physiological coupling decreases, the spatial distribution of the coupling vector tends to diverge, and the corresponding convex hull area also shows an expanding trend. The system identifies the systematic decline of the user's physiological regulation ability by tracking the evolution trajectory of the convex hull area over continuous time periods. When it is detected that the convex hull area continues to expand over a certain period of time and the trend clearly points to a decline in regulatory stability, the present invention will trigger a health risk early warning mechanism. The generation of early warning information does not rely on the abnormality of a single measurement value, but is based on the loss of dynamic stability reflected by the cumulative data of multiple time periods, thereby realizing the transformation of the early warning mechanism from a result-driven to a capability assessment-driven approach.

[0034] In order to further improve the adaptability of the system and reduce the interference of individual differences on the judgment results, the present invention introduces an individual baseline self-calibration mechanism. The system continuously records the user's convex hull area and calculates the statistical average of this indicator over an ultra-long period of 24 to 72 hours, which is used as the individual reference benchmark for the user's physiological coupling regulation stability. In actual monitoring, only when the convex hull area of ​​the current time period is higher than the alarm threshold determined by the benchmark for at least one hour, is it determined that there is a risk of physiological regulation capacity decline, and an early warning message is generated accordingly. In order to enhance the interpretability and intervention guidance significance of the early warning system, the present invention further divides the functional image into two-dimensional coordinate spaces in the coupling vector. The high-response area is the area where the heart rate change rate is 50% higher than the average value in the resting state; the sluggish or disconnected area is the area where the absolute value of the heart rate change rate is lower than 5% of the average value in the resting state. While the coupling vector bundle distribution shows significant morphological diffusion, the system will also count the number of vectors in each quadrant to analyze the pattern characteristics of instability. Specifically, if the number of coupling vectors falling into the high-response area is significantly more than that falling into the sluggish area, it is determined to be hyperactive instability; otherwise, it is determined to be inhibitory instability. The judgment condition of this pattern characteristic is: when the former is greater than the latter multiplied by a preset ratio threshold greater than 1, it is confirmed to be hyperactive; otherwise, it is confirmed to be inhibitory.

[0035] Example 2: This example proposes a health monitoring, analysis and early warning method and system for multimodal data fusion, which aims to use low-power wearable devices to evaluate the early decline of the body's regulatory ability and generate health risk warnings without relying on the absolute values ​​of traditional vital signs. The specific implementation steps are as follows: First, use a wearable device to synchronously collect two physiological data streams: the first physiological data stream is collected by a three-axis accelerometer sensor to reflect the user's micro-movement; the second physiological data stream is obtained by a photoelectric volumetric pulse wave sensor PPG to reflect the user's heart rate changes. The data collection adopts a low-frequency non-uniform wake-up strategy, specifically randomly waking up the device for 5 to 10 seconds every minute for data collection. This strategy can effectively reduce energy consumption and ensure sufficient time resolution and data coverage; after data collection, the motion data stream is first analyzed in the frequency domain. Specifically, the first physiological data stream is processed by short-time Fourier transform (STFT) to determine whether the frequency distribution of motion energy conforms to the physiological low-frequency characteristics. The judgment criterion for physiological low-frequency characteristics is that the proportion of high-frequency energy in the spectrum is lower than the set threshold. Only when this condition is met will the system continue to execute subsequent processing steps. Then, two scalars are calculated based on acceleration data and heart rate data respectively: the motion intensity scalar quantifies the motion intensity by calculating the variance of the acceleration signal within a short time window; the heart rate variability scalar quantifies the heart rate change rate by calculating the slope of the heart rate sequence within a short time window. These two scalars represent the physiological response characteristics of motion and heart rate, respectively. They are then paired to generate a coupling vector. The X-axis of the coupling vector represents the motion intensity, and the Y-axis represents the heart rate variability, which characterizes the physiological coupling response state within the short time window.

[0036] Over a long period of time, the system will accumulate multiple coupling vectors to form a coupling vector cluster, which reflects the user's exercise-heart rate coupling pattern over a long period of time and reveals the inherent law between the user's physiological activity and the response of the autonomic nervous system. In order to evaluate the user's physiological regulation stability, this embodiment measures the discreteness of the coupling vector by calculating the convex hull area formed by the coupling vector cluster in a two-dimensional coordinate system. As the physiological regulation stability decreases, the distribution of the coupling vector tends to diverge and the convex hull area increases. By tracking the evolution of the convex hull area over continuous time periods, the system can identify the decline of the user's physiological regulation ability. When the system detects that the convex hull area continues to increase over a certain period of time and the trend points to a decrease in physiological regulation stability, the system will trigger a health risk warning mechanism. This mechanism is different from the traditional threshold-based alarm system. Instead, it generates warning information based on the dynamic stability changes of multi-period data. In order to further improve the adaptability of the system and reduce the impact of individual differences, this embodiment introduces an individual baseline self-calibration mechanism. The system continuously records the user's convex hull area. The system calculates the average value of the convex hull area over an extended period of 24 to 72 hours. This average serves as an individual benchmark for the stability of the user's physiological coupling regulation. During actual monitoring, the system only determines the risk of physiological regulation decline and generates an alert if the convex hull area of ​​the current time period consistently exceeds the alarm threshold determined by this benchmark. Furthermore, the system divides the two-dimensional coordinate system of the coupling vector into multiple functional quadrants: a high-response region indicates a heart rate variability exceeding 50% of the resting average; a sluggish or disjointed region indicates an absolute heart rate variability less than 5% of the resting average. When the distribution of the coupling vector bundle changes significantly, the system counts the number of coupling vectors in each quadrant and analyzes the instability pattern by comparing the ratio of high-response to sluggish regions. If the number of coupling vectors in the high-response region is significantly greater than that in the sluggish region, it is determined to be hyperactive instability; otherwise, it is determined to be inhibitory instability. This analytical mechanism provides more explanatory information for health risk assessment and offers effective intervention guidance to health managers.

[0037] Example 3: This example aims to verify the actual effectiveness of a multimodal data fusion health monitoring analysis and early warning method and system in evaluating the stability of autonomic nervous system regulation and capturing the early decline of body function through a continuous monitoring test on the physiological response of the human body in daily activities. The core of this test is to achieve objective quantification and trend warning of changes in individual physiological regulation stability by analyzing the dynamic evolution of the exercise-heart rate coupling relationship without relying on the absolute value of traditional vital signs; in the context of the increasing popularity of home health management, how to use low-power, non-invasive wearable devices to monitor the physiological response of the user without or with weak perception. , to achieve a continuous and in-depth assessment of physical health status; existing technologies usually focus on threshold monitoring of the absolute values ​​of physiological parameters. Such methods often lag behind subtle changes in the body's internal regulatory mechanisms, resulting in the lack of early warning windows. To overcome this limitation, this experiment is committed to verifying a new health monitoring, analysis and early warning method. This method converts daily micro-movements of the human body into physiological probes, and by analyzing its dynamic coupling pattern with heart rate, it is expected to achieve real-time assessment of the dynamic stability of the autonomic nervous system's regulatory ability, and thus may capture early signals of body function decline before significant abnormalities occur in conventional physiological indicators, providing users with forward-looking health risk warnings.

[0038] This experiment uses a wearable device that integrates a three-axis accelerometer sensor and a photoplethysmography sensor (PPG) for data collection. The device has low power consumption and communicates with the local data processing unit via the Bluetooth low-power protocol. To ensure the authenticity and daily applicability of data collection, the experimental environment is set as a relatively free daily home activity scene for the subjects, including but not limited to light housework, walking, sitting and other activities. During the experiment, strenuous exercise or environmental factors that may introduce a large amount of non-physiological vibrations have been excluded to ensure that the physiological low-frequency characteristics of the motion signal are effectively reflected. All data collection and processing are completed on the local processing unit, and its configuration can simulate the actual deployed edge computing environment; the subjects of this experiment are digital wireless Healthy volunteers with known cardiovascular or neurological diseases wore wearable devices for continuous monitoring and continuous data recording. The data collection adopted a low-frequency non-uniform wake-up strategy, that is, the device randomly activated the sensor every minute and continuously collected data for several seconds. This strategy aims to balance data coverage and system energy consumption. According to preliminary verification, it can effectively capture the dynamic coupling relationship between movement and heart rate, while extending the single-charge battery life of the device to meet actual home monitoring needs. The collected raw acceleration signals and PPG signals were first pre-processed before entering the subsequent analysis process. The acceleration signal was processed by bandpass filtering to filter out the DC component and high-frequency noise. The heart rate signal was converted into a heart rate sequence through peak detection and interval analysis.

[0039] In each specific short time window, a short-time Fourier transform is performed on the pre-processed acceleration signal. The key to this stage is to determine whether the frequency distribution of motion energy conforms to the physiological low-frequency characteristics. This method implements screening by analyzing the energy proportion of the high-frequency part of the short-time Fourier transform result: if the energy proportion above a specific frequency in the spectrum is lower than a preset judgment threshold, it is considered that the motion signal in the current window mainly comes from the natural, low-frequency physiological micro-movements of the human body, rather than external interference. This window data can enter the subsequent coupling analysis. The determination of this threshold is based on the technical consensus formed by the analysis of a large amount of daily activity data in this field. It aims to retain effective physiological information to the greatest extent while avoiding non-physiological interference, ensuring the evaluation in complex environments. The reliability of the estimation; for the short time window filtered by the frequency domain, the exercise intensity scalar and the heart rate change rate scalar are calculated respectively. The exercise intensity scalar is defined as the variance of the acceleration signal in the short time window. The variance can effectively quantify the activity level of micro-movements. Its calculation method is simple and easy to implement on low-power hardware. The heart rate change rate scalar is defined as the slope of the heart rate sequence in the same time window. The heart rate slope can reflect the immediate rising or falling trend of the heart rate and its rate. It is a direct reflection of the rapid regulatory response of the autonomic nervous system. The determination of these two scalars is based on the profound understanding of the characteristics of physiological signals in this field and the effectiveness considerations in engineering practice. The above-mentioned exercise intensity scalar and heart rate change rate scalar are paired to form a scalar that characterizes the physiological coupling in the short time window. The X-axis of the coupling vector represents the exercise intensity, and the Y-axis represents the heart rate change rate. Over a continuous long period of time, the system accumulates a series of such coupling vectors to form a coupling vector bundle. In order to quantify the morphological discreteness of the coupling vector bundle, this method calculates the convex hull area formed by the coupling vector bundle in the two-dimensional coordinate system. The larger the convex hull area, the more divergent the distribution of the coupling vector, which in turn reflects the decline in the stability of the user's physiological coupling regulation. As a geometric algorithm, the convex hull algorithm has high computational efficiency and can accurately capture the outer boundary of the point set, thereby effectively quantifying its discreteness. In order to adapt to individual differences, this method introduces a personal baseline self-calibration mechanism. The system continuously records the user's convex hull area and calculates the value of the convex hull area within 24 hours to 72 hours. The statistical average of this indicator is calculated over an extremely long period of time, and this statistical average is used as a personal benchmark for judging the stability of physiological coupling regulation. In actual monitoring, only when the convex hull area of ​​the current time period is higher than the alarm threshold determined by the personal benchmark for at least one hour is the risk of physiological regulation decline determined, and a warning message is generated accordingly. This dynamic threshold setting fully considers daily physiological fluctuations and avoids the limitations of rigid fixed thresholds, which would cause the warning logic to be misplaced in the instability of the individual's own regulatory mode. To provide more instructive warning information, this method divides the two-dimensional coordinate system of the coupled vector bundle into the following functional quadrants: a high-response area, where the heart rate change rate is higher than 50% of its average value in the resting state;a dull or disconnected area, in which the absolute value of the heart rate change rate is less than five percent of its average value in the resting state. When it is judged that the stability of physiological coupling regulation shows a declining trend, it further includes counting the number of coupling vectors falling into the high response area; and the number of coupling vectors that fall into the sluggish or disjointed regions , by comparison and The relative size of the physiological coupling regulation instability is used to determine the pattern characteristics. The judgment of the pattern characteristics is based on the following criteria: , it is determined to be hyperactive instability; if , it is determined to be an inhibitory instability, where The preset ratio threshold is greater than 1. This mechanism reuses the existing data structure and can achieve initial screening of pathological tendencies only through partition counting, providing an actionable guidance dimension for family health management.

[0040] During daily activities, the convex hull area of ​​the subjects' coupled vector bundles exhibited regular dynamic changes. During periods of stable emotions and good physiological conditions, the convex hull area was generally small and had a small fluctuation amplitude. The coupled vector bundle had a compact shape, indicating that the coupling relationship between exercise and heart rate was stable and efficient. When the subjects experienced mild physiological load or fatigue, a reversible increase in the convex hull area was observed, but it usually returned to baseline levels within a relatively short period of time, reflecting the normal regulation and adaptability of the physiological system. Through continuous monitoring, it was found that when some subjects showed signs of sub-health in their physiological state, such as feeling mild discomfort but not meeting the diagnostic criteria for a disease, even if the absolute values ​​of their heart rate and acceleration remained within the conventional indicator range, the convex hull area data recorded by the system during the continuous time period showed that their convex hull area remained stable above the personal baseline alarm threshold. This verified that this method can capture the early decline of the body's regulatory stability by expanding the convex hull area before traditional indicators become abnormal. During the experiment, when the subjects were in a state that might introduce non-physiological vibrations, In complex environments, such as when operating certain household appliances, frequency domain analysis revealed that vibration energy is primarily concentrated in the high-frequency range. Thanks to the screening mechanism of physiological low-frequency characteristics, when the high-frequency energy proportion of the acceleration signal exceeds a preset threshold, the data in this short time window is effectively eliminated and does not participate in the subsequent coupling vector generation. This demonstrates that the frequency domain screening mechanism can effectively filter out non-physiological interference, ensuring the purity and reliability of stability assessment results in complex environments. When the subject's physiological coupling regulation stability is judged to be declining, we further analyzed the quadrant distribution of their coupling vector bundle. The results showed that during the period of continuous warning of the convex hull area, there was a significant difference between the number of coupling vectors falling in the high-response region and the number falling in the dull or disconnected region, meeting the criteria for determining hyperactive or inhibited instability. For example, if the number of coupling vectors in the high-response region is significantly greater than that in the dull region, it is judged to be hyperactive instability. The identification of this specific pattern provides more precise guidance for possible subsequent health management interventions.

[0041] Example 4: This example combines Figures 1 to 3 , a health monitoring analysis and early warning method and system based on multimodal data fusion is described. Figure 1As shown in the figure, the data acquisition layer includes two sensor devices: a three-axis accelerometer and a PPG sensor, which synchronously collect motion data streams and heart rate data streams. The data acquisition process is carried out in a low-frequency non-uniform manner of 5 to 10 seconds / minute. The subsequent signal processing layer includes a frequency domain analysis module and a coupling generator. The frequency domain analysis module performs frequency domain analysis on the motion data stream and filters out signals that meet the physiological low-frequency characteristics. Only when the signal meets the conditions, the coupling generator generates a motion intensity scalar and a heart rate change rate scalar based on the motion data stream and the heart rate data stream respectively. These scalars will be paired to form coupling vectors of physiological coupling response. Finally, in the intelligent analysis layer, these coupling vectors will be sent to the vector cluster builder and stability evaluator. The vector cluster builder accumulates multiple coupling vectors over a long period of time to form a coupling vector cluster, while the stability evaluator analyzes the morphological discreteness of the coupling vector cluster and judges the stability of physiological coupling regulation. When the stability shows a declining trend, the system will generate health risk warning information.

[0042] like Figure 2 As shown in the figure, the horizontal axis represents exercise intensity (acceleration variance), the vertical axis represents heart rate variability (bpm / s), and different points represent measurement results in different time periods. The figure uses three different symbols to distinguish the monitoring time periods: the circle represents the measurement results in the afternoon, the square represents the measurement results in the morning, and the triangle represents the measurement results in the evening. It can be seen that with the increase in exercise intensity, the heart rate variability gradually increases, and the measurement data in different time periods show a similar relationship in the overall trend.

[0043] like Figure 3 As shown in the figure, first of all, the data acquisition function module mainly includes pseudo-wearing device acquisition and synchronous acquisition of motion and heart rate data, among which pseudo-wearing device acquisition supports low-frequency mode synchronization and dual-mode synchronization. In the data management function module after data acquisition, medical staff can manage and adjust data through historical data query, baseline parameter setting and warning threshold adjustment functions; the next intelligent analysis function module includes physiological feature extraction and coupling relationship analysis, among which physiological feature extraction can be achieved through real-time health assessment, and coupling relationship analysis is used to analyze the user's motion and heart rate changes, and generate personalized health risk warnings in combination with real-time health assessments. At the same time, under the warning management function module, through risk warning generation and error pattern recognition, the system can generate corresponding warning information based on the predictive analysis results, and push suggestions to help users respond to health risks in a timely manner. In addition, the system also has extended functions, including the extension of stability assessment and individual benchmark calibration, which further improves the accuracy and adaptability of health monitoring.

[0044] Example 5: This embodiment provides a health monitoring, analysis and early warning method and system with multimodal data fusion. To meet the needs of monitoring the physiological status of users in non-wearable or low-perception environments, the system is equipped with an integrated perception module that integrates a three-axis acceleration sensor unit and a photoelectric volume pulse wave detection unit, and periodically exchanges data with the main control unit through a low-power Bluetooth communication protocol. During normal operation, the system wakes up the acquisition module pseudo-randomly at a frequency of once per minute, and the sampling time for each time is set to 6 seconds to 10 seconds, preferably 8 seconds. The sampled data is processed by short-time Fourier transform, using a symmetric Hanning window function, the spectrum resolution is set to 0.5Hz, and the window overlap rate is set to 50% to control computing resource consumption while ensuring the accuracy of frequency separation; and the acquired acceleration modulus data first enters the frequency domain screening link, and the system obtains the spectrum power density distribution of the modulus signal by short-time Fourier transform, and then calculates the energy proportion of the frequency band above 2Hz. When the proportion is less than 30%, it is determined that the motion signal within the sampling period is mainly derived from the user's natural physiological activities. The above 2Hz threshold is based on the following: the frequency of daily human activities is mainly concentrated between 0.5Hz and 1Hz, and signals above 2Hz are usually related to non-autonomous disturbances, such as environmental vibrations or mechanical vibrations, and can therefore be used as noise removal. In addition to the basis, the data filtered by the frequency domain is used to construct a coupling vector. The system calculates the variance of the acceleration modulus in the current sampling window as an indicator of exercise intensity. At the same time, the pulse wave peak interval sequence is extracted and its linear slope is calculated as an indicator of the heart rate change rate. The above two indicators form a two-dimensional coupling vector, where exercise intensity represents local exercise activity, and the heart rate change rate reflects the immediate response trend of the autonomic nervous system to exercise stimulation. The system saves the coupling vector constructed by continuous sampling in a sliding window for long-term trend analysis. The sliding window is set to 15 minutes, contains no less than 20 coupling vector points, and is refreshed with new samples. In this window, all coupling vectors are mapped to a two-dimensional coordinate system with exercise intensity as the horizontal axis and heart rate change rate as the vertical axis. The convex hull calculation module is called to extract the envelope area and calculate its area. The obtained area is used to measure the discreteness of the physiological response pattern during this period.

[0045] In order to take into account individual differences, the system sets up an automatic resting state baseline calibration mechanism. Specifically, within 72 hours of continuous operation after the device is deployed, the system counts the convex hull area every hour and takes the median value as the individual discreteness benchmark of the user. In the subsequent operation process, if the convex hull area in three consecutive sliding windows exceeds 150% of the benchmark value, the system will mark it as a trend of declining regulatory ability and trigger an early warning. Taking a typical application scenario as an example, the user is in a low-intensity activity state for three consecutive hours, such as slow walking or light housework. If the trend of his heart rate change during this period is stable at a level close to the resting level, the relevant coupling vectors will be concentrated in the low-response quadrant. The system further determines that the number of vectors falling into this quadrant is 13, and the high-response quadrant is 13. The number of quadrant vectors should be 3. If the currently set risk judgment threshold is 3, then because 13 is greater than 3 multiplied by 3, the inhibitory instability feature is satisfied. The system will judge that the decline in regulation ability and the abnormal response mode co-occur, thereby generating a high-level risk warning. It should be noted that the frequency domain screening rules adopted can effectively eliminate interference signals such as TV audio, elevator operation and floor vibration in common living environments, thereby improving the validity of physiological data. At the same time, the calculation process involved in this embodiment only relies on conventional algorithms such as basic Fourier transform, variance calculation and convex hull construction, and has good adaptability to software and hardware deployment. It can also operate stably under the condition of limited resources of the embedded main control chip, and can achieve long-term continuous operation in button battery power supply mode.

[0046] Example 6: In the scenario of health monitoring for a 70-year-old user who lives alone for a long time, the multimodal data fusion and early warning mechanism in the system are specifically deployed and executed. The user's wearable wristband device integrates a three-axis accelerometer and a photoelectric volume pulse wave sensor, which works in a low-frequency non-uniform manner. The device main control unit randomly selects a time point every minute to wake up the sensor and synchronously collect eight seconds of acceleration and pulse wave raw data streams, and then enters a dormant state. The acceleration data collected in this process first enters a frequency domain fingerprint identification process. The system performs a short-time Fourier transform on the eight-second acceleration modulus signal to calculate its frequency above a specific frequency. The ratio of the total energy of the (2 Hz) frequency band to the total energy of the signal, the frequency threshold The setting of the function is based on the general rule obtained after statistical analysis of a large amount of physiological activity data: the energy of human body activities, such as walking, getting up, and housework, is mainly concentrated in the low-frequency range below 2 Hz, while vibrations above this frequency are mostly caused by environmental noise or device jitter. Only when the proportion of high-frequency energy is lower than a preset judgment threshold Only when the data segment reaches 30% (30%), the data segment is confirmed as a valid physiological signal and can enter the subsequent analysis link, thus ensuring the purity of the data at the source of the algorithm.

[0047] For the data window that has passed the frequency domain screening, the system then calculates two core scalars, one of which is the motion intensity scalar, by calculating the variance of the three-axis acceleration composite signal within the window. It is concluded that this indicator can sensitively quantify the user's micro-exercise activity. The second is the heart rate variability scalar. The system first extracts the second-by-second heart rate value from the synchronously collected pulse wave data, and then calculates the linear fitting slope of the heart rate sequence within an eight-second window. , the slope directly reflects the dynamic change trend of heart rate in a short period of time. The two scalars are then paired to form a two-dimensional coupling vector This vector accurately depicts the heart rate regulation response of the user's autonomic nervous system under specific micro-motion perturbations. The system continuously accumulates these coupling vectors within a 15-minute sliding time window to form a coupling vector bundle containing dozens of data points. To quantify the morphological discreteness of this vector bundle, the system calls the geometric calculation module to construct and calculate the minimum convex hull area of ​​these data points in the two-dimensional coordinate system. , this area The size of the area directly corresponds to the consistency and stability of the user's exercise and heart rate coupling response pattern within the fifteen minutes: the smaller the area, the more stable and regular the response pattern; the larger the area, the more divergent and unbalanced the response pattern.

[0048] To establish a personalized health assessment baseline, the system performs a personal baseline self-calibration step within seventy-two hours of initial deployment. During this time, the system records all calculated fifteen-minute convex hull areas. The values ​​are statistically sorted and the 75th percentile is taken as the personalized stability benchmark for the user. This benchmark represents a relatively loose but healthy upper limit of the user's ability to adjust to daily physiological fluctuations. Based on this individual benchmark, the system sets a dynamic alarm threshold. , and the relationship is: After completing the baseline calibration and entering the long-term monitoring phase, the system continuously calculates the convex hull area of ​​the current fifteen-minute window. The triggering of the warning is not based on a single area exceeding the standard, but on a clearly defined stable decay trend. The system calculates the moving average of the convex hull area of ​​four consecutive 15-minute windows in the past hour. , only when the moving average Continuously exceeding the alarm threshold Only when the preset duration of one hour is reached will the system determine that the user's physiological coupling regulation stability has experienced a clinically significant and persistent decline and formally generate an early warning.

[0049] Once the physiological coupling regulation stability shows a declining trend, the system will immediately start the qualitative analysis program of the instability mode. The system will project all the coupling vectors accumulated within two hours before the warning is triggered into a two-dimensional coordinate system with clear functional quadrants. The coordinate system is divided according to physiological significance: heart rate change rate The area where the variability of the user's average heart rate is higher than 50% of the resting state is defined as the high response area; Areas with absolute values ​​lower than 5% of the average heart rate change rate at rest are defined as sluggish or disconnected areas. The system counts the number of vectors falling into the high response area. and the number of vectors that fall into the sluggish or disjointed region , by comparing the relative sizes of the two to identify the instability mode, the judgment rule is: if , it is determined to be hyperactive instability; on the contrary, if , it is determined to be an inhibitory instability. The proportional threshold It is set to 3.0, a value based on statistical data from a large clinical population. It is believed that when the number of one response mode is more than three times that of another mode, this imbalance constitutes a significant pathological tendency that requires attention. Ultimately, a detailed health report containing a physiological coupling regulation stability decline alarm, a qualitative conclusion of the instability mode, and relevant data trend charts will be automatically generated and pushed to the user and his or her designated guardian, thus realizing a full-process closed loop from data collection, intelligent analysis to precise warning and in-depth interpretation.

[0050] And, in the specific implementation, when the system determines that the user's physiological coupling regulation stability shows a decline trend, that is, the moving average of the convex hull area for one consecutive hour Continuously above the individual alarm threshold When an alarm is triggered, the system does not simply output a general instability alarm, but automatically activates an in-depth qualitative analysis process to reveal the specific pathological pattern behind the stability degradation. The core of this process is to carefully interpret the distribution pattern of all coupling vectors accumulated in a specific time period (such as two hours) before the alarm is triggered in its two-dimensional coordinate space.

[0051] In the division of this two-dimensional coordinate system, the high response area is defined as the heart rate change rate The vectors in this area represent the over-activation of the sympathetic nerves in the autonomic nervous system, that is, the heart responds excessively to small movements, which is a direct projection of the body being in a regulatory or hyperactive state. In contrast, the sluggish or disjointed area is defined as the heart rate variability. The absolute value of the heart rate variation rate is lower than 5% of the user's average resting heart rate. The vectors in this area reveal that the heart rate regulation is unresponsive or significantly delayed to motion disturbances, which usually points to abnormal vagal tone or obstruction of the cardiopulmonary regulation pathway, and is a sign of fatigue or inhibition of the body.

[0052] To illustrate the value of this qualitative analysis, we take the aforementioned 70-year-old user as an example and observe him in two different time periods. In the first month of monitoring, the system found that his convex hull area It starts to increase continuously and triggers an early warning. In the subsequent qualitative analysis, the system counts the number of vectors that fall into the sluggish or disconnected areas. is 18, and the number of vectors falling into the high response area There are only 2, according to the preset judgment rules, The system accurately classified this instability event as inhibitory instability. The significance of this classification result goes far beyond a simple alarm: it strongly suggests that the user's physiological state is tending towards decreased regulatory function and slow response, which may be a sign of age-related functional decline, chronic fatigue accumulation, or even early cardiac autonomic neuropathy. The health report generated by the system will clearly point out this inhibitory feature and recommend that guardians pay attention to the user's energy state and consider relevant examinations of cardiac function or neural regulatory pathways.

[0053] In the second month of monitoring, the convex hull area of ​​the user Once again, the threshold is continuously exceeded. However, this time the qualitative analysis shows a completely different picture: the system counts the number of vectors that fall into the high response area. is 21, while the number of vectors falling into the sluggish or disjointed region is is 4, because The system classified this instability as hyperactive instability. This conclusion points to a completely different physiological state from the previous one, that is, the body may be in an unstable state caused by high regulation, high anxiety or poor sleep quality, causing it to overreact to small disturbances. Based on this classification, the health recommendations pushed by the system will focus on stress management, improving the sleep environment, or reviewing whether there has been recent use of drugs that affect the nervous system.

[0054] Through the above approach, the present invention successfully interprets a single, quantitative indicator of decreased stability (increase in convex hull area) into two pattern characteristics (inhibitory and hyperactive) with distinct pathological indications and intervention recommendations. This deepening of the problem from discovery to qualitative analysis makes the early warning information no longer a vague signal, but an understandable and actionable early health insight and warning with preliminary diagnostic value.

[0055] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A health monitoring, analysis and early warning method based on multimodal data fusion, characterized in that: The method comprises the following steps: Step a: synchronously collecting a first physiological data stream and a second physiological data stream through a wearable device, wherein the first physiological data stream includes user motion information and the second physiological data stream includes user heart rate information, and the collection is performed in a low-frequency non-uniform manner; Step b: performing frequency domain analysis on the first physiological data stream within a specific short time window to determine whether the frequency distribution of its motion energy conforms to the physiological low-frequency characteristics; Step c: determining an exercise intensity scalar based on the first physiological data stream and a heart rate variability scalar based on the second physiological data stream only when the frequency distribution of the exercise energy conforms to the physiological low-frequency characteristics; Step d, pairing the exercise intensity scalar and the heart rate variability scalar to form a coupling vector that characterizes the physiological coupling response within the short time window; Step e: accumulating a series of coupling vectors over a specific long period of time to form a coupling vector cluster, where the coupling vector cluster reflects the intrinsic pattern of the user's physiological activity and heart rate response; Step f, evaluating the morphological discreteness of the coupling vector bundle, and judging the stability of the user's physiological coupling regulation based on the changing trend of the morphological discreteness in continuous time periods; Step g: When the physiological coupling regulation stability shows a decline trend, health risk warning information is generated.

2. The multimodal data fusion health monitoring, analysis and early warning method according to claim 1 is characterized in that: In step a, the first physiological data stream is acquired by a three-axis accelerometer sensor, and the second physiological data stream is acquired by a photoplethysmography sensor.

3. The health monitoring, analysis and early warning method of multimodal data fusion according to claim 1 is characterized in that: In step c, the exercise intensity scalar is the variance of the acceleration signal in a specific short time window, and the heart rate variability scalar is the slope of the heart rate sequence in the specific short time window.

4. The health monitoring, analysis and early warning method of multimodal data fusion according to claim 1 is characterized in that: In step b, the frequency domain analysis is specifically to perform a short-time Fourier transform on the first physiological data stream, and determine whether the energy proportion of the high-frequency part in the short-time Fourier transform result is lower than a specific judgment threshold; in step f, the method for evaluating the morphological discreteness of the coupled vector bundle is specifically to calculate the convex hull area of ​​the coupled vector bundle in the two-dimensional coordinate system.

5. The health monitoring, analysis and early warning method of multimodal data fusion according to claim 4 is characterized in that: The method also includes a personal baseline self-calibration step: continuously recording the convex hull area and calculating its statistical average over an ultra-long period of twenty-four to seventy-two hours, and using the statistical average as a personal benchmark for judging the stability of physiological coupling regulation.

6. The health monitoring, analysis and early warning method of multimodal data fusion according to claim 5 is characterized in that: In step g, the generation of the warning information is triggered when the convex hull area of ​​the current time window is higher than the alarm threshold determined by the personal benchmark for at least one hour.

7. The health monitoring, analysis and early warning method of multimodal data fusion according to claim 1 is characterized in that: In the two-dimensional coordinate system of the coupled vector bundle, the X-axis represents exercise intensity and the Y-axis represents heart rate variability, which is divided into the following functional quadrants: a high-response area, in which the heart rate variability is higher than 50% of its average value in the resting state; and a sluggish or disjointed area, in which the absolute value of the heart rate variability is less than 5% of its average value in the resting state.

8. The multimodal data fusion health monitoring, analysis and early warning method according to claim 7, characterized in that: In step g, when it is determined that the physiological coupling regulation stability shows a decline trend, the step further includes: counting the number of coupling vectors falling into the high response area and the number of coupling vectors that fall into the sluggish or disjointed regions By comparison and The relative size of the physiological coupling regulation instability is used to determine the pattern characteristics. The judgment of the pattern characteristics is based on the following criteria: , it is determined to be hyperactive instability; if , it is determined to be an inhibitory instability, where is a preset ratio threshold greater than 1.

9. The health monitoring, analysis and early warning method of multimodal data fusion according to claim 1 is characterized in that: The data is collected in a low-frequency, non-uniform manner, specifically by randomly waking up the device for five to ten seconds every minute to collect data.

10. A multimodal data fusion health monitoring, analysis and early warning system, implementing the multimodal data fusion health monitoring, analysis and early warning method according to claims 1-9, the system comprising: A data acquisition module configured to synchronously acquire a first physiological data stream of user motion information and a second physiological data stream of user heart rate information, wherein the acquisition is performed in a low-frequency non-uniform manner; a frequency analysis module configured to perform frequency domain analysis on the first physiological data stream within a specific short time window to determine whether the frequency distribution of its motion energy conforms to the physiological low-frequency characteristics; a coupling vector generation module configured to determine an exercise intensity scalar based on the first physiological data stream and a heart rate variability scalar based on the second physiological data stream only when the frequency distribution of the exercise energy conforms to the physiological low-frequency characteristics, and to pair the exercise intensity scalar and the heart rate variability scalar to form a coupling vector representing the physiological coupling response within the short time window; A coupling vector cluster building module is configured to accumulate a series of coupling vectors over a specific long period of time to form a coupling vector cluster, where the coupling vector cluster reflects an intrinsic pattern of the user's physiological activity and heart rate response; a stability evaluation module configured to evaluate the morphological discreteness of the coupling vector bundle and determine the stability of the user's physiological coupling regulation based on a changing trend of the morphological discreteness within a continuous time period; The early warning generation module is configured to generate health risk early warning information when the physiological coupling regulation stability shows a decline trend.

Images