Health monitoring system based on multi-domain feature extraction of pulse signal and deep learning

Abstract

The application discloses a health monitoring system based on pulse signal multi-domain feature extraction and deep learning, wherein a multi-dimensional feature extraction module adopts an FMCW radar and a sensor device to collect and acquire pulse signals, and carries out multi-level processing and feature extraction on the acquired pulse signals to extract features related to a health state; a feature correlation mapping module matches and correlates different frequency band pulse signal features with the health state; a state recognition module automatically recognizes and classifies the pulse signal features and the health state based on a deep learning model; and a health information feedback module analyzes the user's health state and provides health feedback according to the health state recognition result of the state recognition module, and generates corresponding personalized health guidance information. The system is beneficial to comprehensively analyzing the health state, providing continuous and targeted health tracking and evaluation services, and meeting the modern individualized intelligent Chinese medicine health monitoring and management requirements.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traditional Chinese medicine health monitoring and artificial intelligence technology, in particular to a health monitoring system based on pulse signal multi-domain feature extraction and deep learning for assisting traditional Chinese medicine health analysis and evaluation. BACKGROUND

[0002] At present, as an important part of the four diagnostic methods of traditional Chinese medicine, pulse diagnosis mainly relies on the experience and fingertip feeling of traditional Chinese medicine practitioners to judge and evaluate the health status of patients in traditional practice. However, due to subjective factors and lack of standardization, the accuracy and repeatability of traditional pulse diagnosis are limited. In addition, most of the existing instrument systems rely on time domain analysis, which can only extract the surface features of the pulse waveform, and cannot deeply analyze the frequency components, phase changes and other information. Moreover, it does not combine the systematic scientific thinking of amplification feedback, resulting in insufficient depth and comprehensiveness in interpreting the health status of viscera, and it is difficult to meet the fine monitoring needs of the functional status of viscera, and it is also difficult to accurately identify the trend of health status change. Moreover, it restricts the improvement of the efficiency of integrated traditional Chinese and Western medicine diagnosis and treatment. The current technology lacks the ability to adapt to individual differences, and it is difficult to conduct personalized health analysis according to the unique physiological characteristics of users. This limits the application of the technology in personalized health management, making it difficult to provide continuous and targeted health tracking and evaluation services, and unable to meet the modern individualized intelligent health monitoring and management needs. SUMMARY

[0003] Therefore, the present application provides a health monitoring system based on pulse signal multi-domain feature extraction and deep learning, which overcomes the limitations of the prior art by using pulse signal multi-domain feature extraction and deep learning technology, provides a scientific basis for health evaluation, realizes accurate and comprehensive analysis of health status, and provides continuous and targeted health tracking and evaluation services, which is conducive to meeting the modern individualized intelligent health monitoring and management needs of traditional Chinese medicine.

[0004] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0005] In a first aspect, the present application provides a health monitoring system based on pulse signal multi-domain feature extraction and deep learning, which comprises a multi-dimensional feature extraction module, a feature correlation mapping module, a state recognition module and a health information feedback module, wherein:

[0006] The multi-dimensional feature extraction module uses FMCW radar and sensor equipment to collect and acquire pulse signals, and performs multi-level processing and feature extraction on the acquired pulse signals to extract features related to health status;

[0007] The feature correlation mapping module is configured to match and correlate different frequency band pulse signal features with health states.

[0008] The state recognition module is configured to automatically recognize and classify pulse signal features and health states based on a deep learning model.

[0009] The health information feedback module is configured to analyze user health states and provide health feedback based on the health state recognition results of the state recognition module, and generate corresponding personalized health guidance information.

[0010] In an optional embodiment, the system further comprises a state tracking and trend analysis module configured to perform time series analysis based on pulse signal data and draw a health trend graph that changes over time.

[0011] In an optional embodiment, the system further comprises a personalized baseline measurement module configured to establish an initial reference baseline of a user's health state, and to prioritize comparison with the initial reference baseline when analyzing each detection result of the user to determine abnormal health changes.

[0012] In an optional embodiment, the system further comprises an AR display module configured to use augmented reality technology to present information collected by the FMCW radar and sensor device, health feedback provided by the health information feedback module and generated personalized health guidance information, and a health trend graph drawn by the state tracking and trend analysis module on a real-world view.

[0013] In an optional embodiment, the operating frequency range of the FMCW radar covers 24 GHz to 77 GHz; the sensor device comprises an infrared sensor and a pressure sensor, wherein the infrared sensor collects optical waveform information of the pulse by detecting changes in light absorption caused by blood flow; and the pressure sensor directly captures pressure changes caused by pulse fluctuations through mechanical contact.

[0014] In an optional embodiment, the multi-dimensional feature extraction module performs multi-level processing and feature extraction on the acquired pulse signal, specifically including time domain analysis, frequency domain analysis, and nonlinear feature analysis, wherein:

[0015] In time domain analysis, basic features of the pulse signal are extracted, including the amplitude, peak, trough, and pulse transmission time of the waveform.

[0016] In frequency domain analysis, the pulse signal is converted from the time domain to the frequency domain, and frequency decomposition is performed to extract the energy distribution of different frequency bands.

[0017] In nonlinear feature analysis, the dynamic complexity and zang-fu function stability of the pulse signal are evaluated using entropy analysis and Lyapunov exponent methods.

[0018] In an optional embodiment, in the frequency domain analysis, the pulse signal is first converted into frequency domain information by Fourier transform to obtain the energy distribution of the signal at different frequencies; then the time domain signal is restored by inverse Fourier transform, and the analytic signal is generated using Hilbert transform; finally, the analytic signal is combined with the original signal to obtain a signal containing amplitude and phase information; the calculation formula includes:

[0019]

[0020] In the formula, X(f) represents the signal spectrum of the time domain signal x(t) after Fourier transform; f represents the frequency variable; t represents the time variable, which is the time position of the signal in the time domain; j represents the imaginary unit; e represents the complex exponential function; X n (t) represents the frequency domain information obtained by Fourier transform of the signal x n (t), that is, the frequency spectrum representation of the signal, x n (t) represents the pulse signal obtained after preprocessing the original pulse signal x(t); represents the Hilbert transform to generate the analytic signal, τ represents the time delay or lag, which is a virtual integral variable in the Hilbert transform process in this formula; z(t) represents the signal obtained by combining the analytic signal with the signal x(t); A(t) represents the instantaneous amplitude information of the signal; φ(t) represents the instantaneous phase information of the signal.

[0021] In an optional embodiment, the feature correlation mapping module includes a frequency domain and viscera function correlation mapping unit, an amplitude and fluctuation amplitude analysis unit, and a waveform feature database, wherein:

[0022] The frequency domain and viscera function correlation mapping unit is configured to correlate the pulse signal features of different frequency bands with the viscera function state, and analyze the viscera state according to the fluctuation of a specific frequency;

[0023] The amplitude and fluctuation amplitude analysis unit is configured to analyze the envelope of the pulse signal after Hilbert transform to obtain the instantaneous amplitude and phase information, determine the strength of the viscera function through the fluctuation amplitude change of the amplitude, and observe the flow state of the blood and qi through the dynamic change of the phase;

[0024] The waveform feature database is configured to store pulse diagnosis data and pulse condition features, and provide pattern references for viscera function state judgment; the database contains standard pulse signal waveform data under various constitutions and health states, and has diversified blood and qi and viscera state patterns.

[0025] In an optional embodiment, the state recognition module uses a ResNet101 deep learning model to realize automatic recognition and classification of pulse signal features and health states through layer-by-layer convolution.

[0026] In an optional embodiment, when analyzing the user's health state, the health information feedback module also comprehensively evaluates the user's pulse information, respiratory rate and body temperature information, and the evaluation method is:

[0027] Q = a0 + a1|F - F'| + a2|B - B'| + a3|T - T'

[0028] In the formula, Q is the health index, representing the health state value after comprehensive evaluation; a0 is a constant term, representing the basic health index when there is no any deviation; a1 represents the weight coefficient of pulse deviation; F represents the actual pulse value, and F' represents the standard pulse value; a2 represents the weight coefficient of respiratory rate deviation, B represents the actual respiratory rate, and B' represents the standard respiratory rate value under the ideal health state; a3 represents the weight coefficient of body temperature deviation, T represents the current body temperature value, and T' represents the standard body temperature value.

[0029] Compared with the prior art, the present application provides a health monitoring system based on pulse signal multi-domain feature extraction and deep learning, which overcomes the limitations of the prior art through multi-domain feature extraction and machine learning technology.

[0030] More comprehensive information acquisition: through the combination of time domain, frequency domain and short-time transformation, the system can not only capture the intensity information of the pulse wave, but also extract the instantaneous energy distribution and phase change of the viscera, making the health monitoring analysis more comprehensive, in-depth and accurate.

[0031] Innovative multi-domain signal processing method: using multiple combinations of wavelet, Fourier and Hilbert transform, a multi-dimensional and accurate pulse signal analysis system is formed, which has significant innovation in multi-domain signal processing.

[0032] Accurate visceral health association: using frequency decomposition and feature mapping, the pulse signal is accurately associated with the health status of the five internal organs and six viscera and the TCM syndrome elements, which facilitates to provide more targeted health evaluation results.

[0033] Automation and personalization: based on the machine learning model, the system realizes automatic recognition of health status and can be adjusted according to the user's personalized baseline, which is convenient for adapting to the health needs of different users.

[0034] Dynamic monitoring and health feedback: the system tracks the user's health status for a long time and provides timely warning when the health fluctuates, which is convenient for providing dynamic health monitoring management and feedback support for users.

[0035] Intelligence: the system adopts AR augmented reality, can present information directly on the real world view, improves the intelligence of traditional Chinese medicine health monitoring, is more intuitive and efficient, and is beneficial to meet the modern individualized intelligent traditional Chinese medicine health monitoring management demand.

[0036] Other features and advantages of the present application will be set forth in the description that follows, and in part will become apparent to those skilled in the art upon examination of the following or can be learned by practice of the application. The objects and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

[0037] The technical solutions of the present application will be further described in detail below by means of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0039] The accompanying drawings are used to provide a further understanding of the present application, and constitute a part of the specification, together with the embodiments of the present application, to explain the present application, and do not constitute a limitation on the present application.

[0040] Figure 1 A health monitoring system structure schematic diagram based on pulse signal multi-domain feature extraction and deep learning is provided for the embodiments of the present application.

[0041] Figure 2 An electronic device structure schematic diagram is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0042] In order to make the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments.

[0043] In the description of the present application, it should be noted that: in some processes described in the present application specification and drawings, a plurality of operations appear in a specific order, but it should be clearly understood that these operations can be executed or executed in parallel without the order appearing in this text. In addition, various serial numbers and the like are only for the purpose of description, and cannot be understood as indicating or implying relative importance.

[0044] The following detailed description of embodiments of the application provided in the accompanying drawings is not intended to limit the scope of the application claimed, but merely represents selected embodiments of the application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0045] Referring to Figure 1 As shown, the present application provides a health monitoring system based on multi-domain feature extraction and deep learning of pulse signals, which uses a multi-dimensional signal processing method to extract subtle features in the pulse signal (such as the fingertip), combines machine learning models and feature correlation mapping, automatically analyzes the health status of the five internal organs and six viscera, and realizes health assessment of the upper, middle and lower Jiaos and extraction of TCM syndrome elements. The system realizes automatic pulse diagnosis analysis and health monitoring evaluation through a combination of software and hardware, and mainly includes a multi-dimensional feature extraction module, a feature correlation mapping module, a state recognition module based on deep learning, and a health information feedback module. In some preferred embodiments, the system further includes a state tracking and trend analysis module, a personalized baseline measurement module, and an AR display module, etc.

[0046] In one specific embodiment, by adopting FMCW frequency-modulated continuous wave technology, the present application overcomes the inconvenience of contact measurement in traditional detection methods, and realizes non-inductive acquisition of pulse signals. FMCW not only can accurately measure the speed change of the pulse wave, but also can realize synchronous ranging, providing reliable technical support for comprehensive acquisition of pulse signals. The present application uses a multi-dimensional signal processing method to extract weak features in the pulse signal to reveal potential health status information. The feature extraction method in non-stationary and nonlinear pulse signals is explored to mine key features that can reflect the functional status of the five internal organs and six viscera, and these features are combined with the amplification feedback theory to realize comprehensive analysis of the user's health status and support fine classification prediction of the health status. The present application uses specific frequency decomposition and signal processing technology to separate different frequency components of the pulse signal, and explores the matching and correlation between each frequency component and the specific function of the viscera. The goal is to identify the frequency band related to the specific function of the five internal organs and six viscera in the pulse signal spectrum, thereby providing a scientific basis for viscera health assessment and realizing the deep integration of pulse signals and the TCM diagnosis and treatment system. The present application improves the intelligent level and evaluation accuracy of pulse signal processing through deep learning algorithms (such as ResNet). Through the application of deep learning networks, personalized health management is supported, and the human health status and the trend of yin and yang changes are accurately monitored, providing technical support for the modernization of TCM diagnosis and treatment.

[0047] The working principles and specific embodiments of each module of the system of the present application are described in detail as follows:

[0048] 1. Multi-dimensional feature extraction module:

[0049] In the embodiments of the present application, the hardware structure adopts a dual-sensor system, which includes a millimeter wave FMCW radar and a fingertip sensor, and the two work together to ensure comprehensive capture of pulse signals and data reliability. The fingertip sensor part includes an infrared sensor (OSRAM SFH 7050) and a pressure sensor (Honeywell FSS-SMT), which complement each other: the infrared sensor collects optical waveform information of the pulse by detecting the light absorption changes caused by blood flow; the pressure sensor directly captures the pressure changes caused by the pulse fluctuation through mechanical contact. This dual-mode data acquisition can more comprehensively reflect the real biological information of the pulse. On the other hand, the millimeter wave FMCW radar realizes non-contact pulse signal detection, avoiding the compression influence or measurement error that may be caused by traditional contact methods. The FMCW radar uses the BGT60 series millimeter wave radar chip of Infineon Company, and the working frequency range covers 24GHz to 77GHz. The higher working frequency significantly improves the detection sensitivity of the millimeter wave radar to small displacements, and it can capture the micron-level micro-motion signals caused by the pulse fluctuation on the surface of the radial artery of the human body.

[0050] The working principle of the FMCW radar is to emit a linear frequency modulation signal, and the frequency of the signal increases or decreases linearly with time. After the signal is irradiated to the target surface, it is reflected back to the receiving end of the radar. The micro-motion caused by the pulse leads to changes in the phase and frequency of the reflected signal, and a beat frequency signal is generated through mixing processing. The frequency difference of the beat frequency signal is related to the target distance and speed information. The specific data acquisition process includes: the radar antenna continuously emits a frequency modulation signal while receiving the echo signal reflected by the skin surface; after mixing the received signal with the transmitted signal, a beat frequency signal containing target distance and micro-motion information is generated. Through fast Fourier transform (FFT), the beat frequency signal is analyzed in the frequency domain, and the distance between the human skin surface and the radar is accurately calculated, and environmental noise interference is eliminated. Subsequently, the phase difference method and I / Q demodulation technology are used to track the phase change of the reflected signal in real time, and the small displacement information of the skin is further extracted, which is converted into time series data, so as to reconstruct the pulse waveform.

[0051] After the data acquisition is completed, the system performs multi-level processing and feature extraction on the acquired pulse signals, including time domain analysis, frequency domain analysis and nonlinear feature analysis. First, in the time domain analysis, the basic features of the pulse signal are extracted, such as the amplitude, peak, valley and pulse transmission time (PTT) of the waveform, which can intuitively reflect the fluctuation form, strength change and dynamics of blood flow, providing basic data for heartbeat rhythm and blood vessel elasticity evaluation.

[0052] In frequency domain analysis, the pulse signal is converted from time domain to frequency domain by Fourier transform, and its frequency components are extracted and analyzed in association with the functions of the five Zang organs. In addition, to further capture the transient dynamic changes in the pulse signal, wavelet transform is used to decompose the signal at multiple levels. Wavelet transform has good short-time analysis capability, which can decompose the pulse signal into detailed features at different time scales and frequency bands, and analyze the dynamic changes from low frequency to high frequency step by step, so as to reveal the details of pulse condition closely related to the functions of the five Zang organs. In frequency domain analysis, the specific calculation formula includes the following:

[0053] The signal is converted from time domain to frequency domain by Fourier transform, and the distribution of the signal at different frequencies is obtained:

[0054]

[0055] In the formula, X(f): Fourier transform of the signal x(t) in the frequency domain, representing the frequency spectrum of the signal. f: frequency variable, representing different frequency components of the signal in the frequency domain. x(t): time domain signal, representing the signal (such as pulse wave, pressure wave, etc.) varying with time. t: time variable, representing the time position of the signal in the time domain. j: imaginary unit; e: complex exponential function, representing the modulation process, through which the changes of the signal at different frequency components can be analyzed.

[0056] Then the signal is decomposed, and in the frequency domain, X(f) is decomposed into different parts according to different frequency components. Through biological experiments, the frequency bands matched with the twelve Zang organs are selected to extract a part of the signal, and then the inverse Fourier transform is performed on these different frequency components to convert them to time domain signals:

[0057]

[0058] In the formula, X n (t) represents the frequency domain information obtained by Fourier transform of the signal x n (t), that is, the frequency spectrum representation of the signal, x n (t) represents a processed version of the signal x(t), usually the pulse signal obtained after time domain analysis, filtering or other preprocessing.

[0059] Then perform Hilbert transform: restore to time domain signal by inverse Fourier transform, use Hilbert transform to generate analytic signal Analyze the instantaneous amplitude and phase changes to further reveal the dynamic health status of the twelve Zang organs:

[0060]

[0061] where τ represents the time delay or lag, which is a virtual integration variable in the Hilbert transform process. This variable is used in the integration process of the signal and the generation of the analytic signal, which describes the relative change of the signal between different time points.

[0062] The analytic signal z(t) is constructed as the complex of the original signal and the Hilbert transform:

[0063]

[0064] where the instantaneous amplitude of the envelope signal is: The instantaneous phase of the signal is:

[0065] In the nonlinear feature analysis, the fuzzy Lyapunov exponent is introduced to solve the noise sensitivity problem in the traditional Lyapunov exponent calculation, and to evaluate the dynamic complexity and system stability of the pulse signal. The calculation formula is:

[0066]

[0067] where λ f(t) is the fuzzy Lyapunov exponent, which represents the local dynamic characteristics of the system at time t. It measures the rate of change of the system state and is used to judge the stability of the system. The larger the Lyapunov exponent, the more unstable the system dynamics; the smaller the Lyapunov exponent, the more stable the system. μ(·) represents the fuzzy membership function, which is used to process the uncertainty of the signal; Δt represents the time window or time step, which controls the time step or time difference in the signal sequence. By changing the size of Δt, the stability of the system at different time scales can be analyzed. |x(t+Δt)-x(t)| represents the difference between the pulse signal at time t+Δt and time t, which measures the degree of change of the pulse signal in this time step. This difference reflects the instantaneous change of the pulse signal. |x(t)-x(t-Δt)| represents the difference between the pulse signal at time t and time t-Δt, which similarly measures the degree of change of the pulse signal. ∑ i represents the summation over all time points i. Here, the summation is performed on each pair of time differences t and t+Δt in the signal sequence.

[0068] Specifically, the complexity of the pulse signal is calculated using approximate entropy and sample entropy to reveal the regularity and randomness of the signal, and to evaluate the coordination degree of the function activities of the viscera, especially the kidney function status. Lyapunov exponent analyzes the nonlinear dynamic characteristics of the pulse wave signal to judge the stability and self-excitation oscillation phenomenon of the viscera function system, providing a basis for deep evaluation of health status.

[0069] Combined with the results of time domain, frequency domain and nonlinear feature analysis, the system can effectively extract the feature indexes closely related to the function activities of five zang organs, and realize multi-dimensional evaluation and dynamic monitoring of the health status of zang-fu organs.

[0070] Finally, in order to ensure the accuracy and reliability of the data, the system synchronously fuses the pulse waveform extracted by the FMCW radar and the infrared and pressure signals collected by the fingertip sensor, and preferably adopts a weighted average feature fusion method to combine the feature information of the pulse waveform extracted by the FMCW radar and the infrared and pressure signals collected by the fingertip sensor, and the formula is:

[0071]

[0072] Wherein: w i is the weight of the feature of each signal source, so as to obtain a more accurate pulse signal feature; n represents the number of signal source features; after feature extraction and fusion processing, a pulse signal f(t) is obtained, which is a comprehensive feature of multiple sensors, and has higher stability and accuracy. The fusion signal can effectively reflect the change characteristics of the pulse signal, and provide reliable data support for subsequent health monitoring and analysis.

[0073] Further, the fusion data is filtered and baseline corrected to remove external interference and noise, and a more stable and accurate pulse signal is obtained.

[0074] 2. Feature correlation mapping module:

[0075] In the embodiment of the application, the feature correlation mapping module is composed of a frequency domain and zang-fu function correlation mapping unit, an amplitude and fluctuation amplitude analysis unit, and a waveform feature database, which aims to reveal the deep relationship between the pulse waveform and the five zang organs and six fu organs through multi-dimensional feature correlation, and the specific functions are as follows:

[0076] The frequency domain and zang-fu function correlation mapping unit: the pulse signal is analyzed in the frequency domain, and the features of different frequency bands are correlated with the function states of five zang organs and six fu organs. According to the theory of traditional Chinese medicine, the characteristics of liver channel, kidney channel, spleen channel, lung channel, etc. are usually more obvious in the low frequency band, while the characteristics of triple energizer channel and small intestine channel are usually in the high frequency band. Through the correlation mapping of this frequency characteristic, the system can analyze the zang-fu state according to the fluctuation of the specific frequency, and provide accurate frequency basis for health evaluation.

[0077] The amplitude and fluctuation amplitude analysis unit: the envelope of the pulse signal after Hilbert transform is analyzed to obtain the instantaneous amplitude and phase information. The system judges the strength of the zang-fu function through the fluctuation amplitude change of the amplitude, and observes the flow state of the blood and the blood through the dynamic change of the phase. This unit can describe the rise and fall of the zang-fu in detail, especially in the dynamic balance of the yin and yang balance of blood and blood, and capture the abnormality, which provides accurate data support for health evaluation.

[0078] Waveform feature database: This database is used to store a large amount of pulse diagnosis data and pulse condition characteristics, providing pattern reference for the diagnosis of five internal organs and six viscera. The database contains standard waveform data under various constitutions and health conditions, and has diversified patterns of qi-blood and zang-fu states. The system can identify abnormal health conditions of the user through matching analysis with these standard waveforms, and generate personalized health recommendations.

[0079] 3. State recognition module based on deep learning:

[0080] In the embodiments of the present application, the state recognition module based on deep learning adopts ResNet101 deep learning algorithm to realize automatic recognition and classification of pulse signal features and health status. The specific implementation is as follows:

[0081] ResNet101 model training unit: This unit uses historical data of multi-dimensional features for model training. The model can identify the health status of different parts such as upper jiao (heart, lung), middle jiao (spleen, stomach), lower jiao (liver, kidney, bladder), and three yin and three yang.

[0082] Automatic recognition and classification unit: The system automatically classifies the input data into corresponding health states and outputs the evaluation results. In addition, based on the mapping relationship between waveforms and syndrome elements, the model associates pulse waveforms and specific TCM syndromes with health status.

[0083] The present application utilizes multi-domain feature data and combines deep learning algorithm to realize accurate identification of health status through layer-by-layer convolution. Based on the multi-dimensional feature mapping of qi-blood and zang-fu function states, the model can identify syndrome elements such as "qi deficiency" and "lung", and further form comprehensive syndromes such as "lung qi deficiency syndrome". In the health monitoring system of the present application, the state recognition module uses a deep learning model based on the theory of three yin and three yang to classify the detected physiological features, in order to accurately evaluate the health status of the human body. Shaoyang Sanjiao and Taishen Pidu form Taiji, respectively dominating the basic temperature and humidity of the human body, and interacting with each other to generate Ying and Wei, blood and spirit, and maintain life activities. Pathologically, Shaoyang yang deficiency leads to the failure of yin essence to ascend, easily causing heart and lung diseases; Taishen spleen deficiency leads to the failure of water and dampness to transform, causing damp diseases. The system identifies states such as Shaoyang yang deficiency and Taishen spleen deficiency through analysis of pulse and health characteristics, and identifies the dynamic balance of jueyin wind-fire and yangming dryness according to the spring yang rising of jueyin liver and the autumn yin sinking of yangming lung, so as to judge the disease trend, such as jue heat dominance or internal injury dryness disease. In addition, the system analyzes the cold and heat confusion at the winter solstice and summer solstice according to the yin and yang extreme transformation rule of the sun and the sun, and identifies the health risks caused by alternating cold and heat.

[0084] 4. Health information feedback module:

[0085] In the embodiment of the present application, the health information feedback module comprises a health feedback unit and a conditioning suggestion unit, wherein:

[0086] The health feedback unit is configured to analyze the visceral state of the user according to the system detection result and provide accurate health feedback. The feedback content is refined to the balance state of qi, blood, yin and yang in the upper, middle and lower jiao regions (heart, lung, spleen, stomach, liver, kidney and bladder), and the specific functional state of the five internal organs and six viscera is differentiated. The feedback content includes common TCM syndromes such as heart yin deficiency, heart yang deficiency, heart qi deficiency, lung qi deficiency, etc., so that the user can intuitively understand the health status of each visceral organ.

[0087] The conditioning suggestion unit generates corresponding personalized health guidance according to the differentiation result of the health feedback unit. The suggestion content is specific to diet, lifestyle and work and rest adjustment, and provides specific conditioning methods for the user according to the health status of different viscera. For example, for users with heart yin deficiency, the system will recommend diet conditioning to nourish heart yin; and for users with weak spleen and stomach, it will suggest diet selection and work and rest adjustment to warm and nourish the spleen and stomach.

[0088] The specific feedback mode is as follows:

[0089] Upper jiao (heart, lung) feedback: The upper jiao is mainly responsible for the distribution of qi. If the heart yang is deficient, it will show that the heart beats weakly and is short of breath, etc. The feedback unit will prompt the user that the heart and lung function is not good, and there may be qi deficiency and yang deficiency. The conditioning suggestion unit will suggest that the user do deep breathing exercises more often, intake foods that nourish the heart and lung (such as lotus seeds and tremella), and appropriately do heart and lung function exercises.

[0090] Middle jiao (spleen, stomach) feedback: The middle jiao is mainly responsible for digestion and transformation. The functional state of the spleen and stomach corresponds to the source of qi and blood in TCM theory. If the spleen qi or stomach qi is detected to be deficient, the system will feedback that the function of the spleen and stomach is weak. The conditioning suggestion will recommend that the user add easily digestible and spleen-strengthening foods (such as yam and coix seed) to the diet, and reduce the intake of cold and raw foods. At the same time, it is suggested to appropriately strengthen exercise and diet.

[0091] Lower jiao (liver, kidney, bladder) feedback: The lower jiao corresponds to the excretion and reproductive function, and is related to the health of kidney yang and kidney essence. If the lower jiao function is abnormal, such as kidney qi deficiency or kidney yin deficiency, the system will feedback that the kidney function is insufficient, which may be accompanied by symptoms such as fatigue and soreness of the waist and knees. The conditioning suggestion will guide the user to choose kidney-tonifying foods (such as black sesame seeds and Chinese wolfberry) and appropriately adjust the work and rest to improve the energy state.

[0092] In the syndrome differentiation feedback process of qi-blood-yin-yang, the health information feedback module refines the health status, and if heart qi deficiency, kidney yin deficiency and other syndromes are detected, accurate syndrome differentiation is provided in combination with the classification model result, and personalized conditioning suggestions are given, such as recommending warming conditioning for heart yang deficiency and recommending nourishing heart for heart yin deficiency, so as to realize individualized traditional Chinese medicine health intervention.

[0093] In an optional embodiment, when the health information feedback module analyzes the user's health status, the user's pulse information, respiratory rate and body temperature information are also combined for comprehensive evaluation, and the specific evaluation calculation is as follows:

[0094] Q = a0 + a1|F-F'| + a2|B-B'| + a3|T-T'|

[0095] In the formula, Q is the health index, which represents the health status value after comprehensive evaluation; a0 is a constant term, which represents the basic health index when there is no any deviation; a1 represents the weight coefficient of pulse deviation; F represents the actual pulse value, and F' represents the standard pulse value; a2 represents the weight coefficient of respiratory rate deviation, B represents the actual respiratory rate, and B' represents the standard respiratory rate under the ideal health state; a3 represents the weight coefficient of body temperature deviation, T represents the current body temperature value, and T' represents the standard body temperature value.

[0096] By quantifying the health status, the health problems are further identified and focused on, and appropriate health intervention measures are guided to be taken; by joint analysis of multiple vital signs, comprehensive evaluation of the health status is realized, so as to provide scientific data support for preventive health management and early disease diagnosis in combination with health trend prediction.

[0097] 5. State tracking and trend analysis module:

[0098] In the embodiment of the application, the state tracking and trend analysis module comprises a time series analysis unit and a health warning unit, wherein:

[0099] Time series analysis unit: by performing time series analysis on the pulse wave data, the system can draw a health trend graph of the function of viscera and bowels changing with time, and can clearly show the changes of the health status of five viscera and six bowels in upper (heart, lung), middle (spleen, stomach) and lower (liver, kidney, bladder) regions. Through the trend graph, the user can observe the fluctuation of the function of viscera and bowels. For example, in the case of long-term stress or irregular work and rest, the heart and lung health may show a downward trend, and the system will intuitively present these changes.

[0100] Health warning unit: The health warning unit generates corresponding reminders based on abnormal situations in the trend chart, such as significant fluctuations or long-term declining trends in the state of a certain Zangfu. Through the warning mechanism, users can pay attention to their health status in a timely manner and take targeted adjustments, such as strengthening exercise or adjusting diet, to prevent further health problems.

[0101] 6. Personalized baseline measurement module:

[0102] In the embodiments of the present application, the personalized baseline measurement module includes a personalized baseline establishment unit and a personalized analysis unit, wherein:

[0103] Personalized baseline establishment unit: This unit allows users to record personal baseline pulse characteristics at the first detection, establishing an individualized initial reference baseline. The baseline includes the user's pulse waveform features and Zangfu state, which will serve as the basis for comparison in subsequent detection, enabling the system to more accurately judge changes in health status. The core of the personalized baseline measurement module is to establish an individualized health baseline for each user to reduce the judgment bias caused by ignoring individual differences in conventional health detection. In this way, the system can more accurately adapt to the unique pulse characteristics of the user, improving the accuracy of health status assessment. In the first use of the user, the system will record the user's pulse data and establish a personalized health baseline. The baseline data includes the basic features of the pulse wave in the time domain and frequency domain, which can reflect the user's overall Zangfu state and adapt to individual pulse characteristics.

[0104] Personalized analysis unit: Based on the initial baseline data of the user, the system can more accurately identify the health status that conforms to the user's personal characteristics. This unit will prioritize comparison with the personalized baseline when analyzing the user's detection results each time, reducing errors caused by individual differences and improving the accuracy of health assessment. Personalized analysis: In each new detection, the system compares pulse characteristics with the baseline. Such comparison can eliminate noise caused by the user's own pulse characteristics, and the system can more quickly and accurately judge abnormal health changes and generate health recommendations that conform to the user's characteristics.

[0105] 7. AR display module:

[0106] In the embodiments of the present application, the AR display module can be AR glasses, computers, or tablet terminal devices; using augmented reality technology, the information collected by the FMCW radar and sensor devices, the health feedback provided by the health information feedback module, the personalized health guidance information generated, and the health trend chart drawn by the state tracking and trend analysis module, etc. Information is presented on the real-world view.

[0107] The module based on AR technology allows users to maintain contact with the real world while providing additional information or functions; making it very suitable for application scenarios that require real-time information overlay; users can obtain a more immersive data presentation experience without completely leaving the real environment; AR can integrate virtual elements into the user's actual environment, providing a new way of participation for traditional Chinese medicine monitoring; enhances the experience in real life without completely leaving the current environment; helps to improve the intelligence of traditional Chinese medicine health monitoring, more intuitive and more efficient; helps to meet the modern individualized intelligent traditional Chinese medicine health monitoring management needs. In addition, VR display modules and conventional liquid crystal display technologies can also be used to present information, which will not be described here.

[0108] From the description of the above embodiments, those skilled in the art can know that the present application proposes a health monitoring system based on pulse signal multi-domain feature extraction and deep learning, which overcomes the limitations of the prior art through multi-domain feature extraction and machine learning technology; the present application has the following advantages:

[0109] 1. More comprehensive information acquisition: through the combination of time domain, frequency domain and short-time transformation, the system can not only capture the intensity information of the pulse wave, but also extract the instantaneous energy distribution and phase change of the viscera, making the health monitoring analysis more comprehensive and in-depth.

[0110] 2. Innovative multi-domain signal processing method: using multiple combinations of wavelet, Fourier and Hilbert transform, a multi-dimensional and accurate pulse signal analysis system is formed, which has significant innovation in multi-domain signal processing.

[0111] 3. Accurate viscera health correlation: using frequency decomposition and feature mapping, the pulse signal is accurately correlated with the health status of the five internal organs and six viscera and the traditional Chinese medicine syndrome elements, which facilitates to provide more targeted health assessment results.

[0112] 4. Automation and individualization: based on the machine learning model, the system realizes automatic identification of health status and can be adjusted according to the user's individualized baseline, which is convenient for different users to adapt to the health needs.

[0113] 5. Dynamic monitoring and health feedback: the system tracks the user's health status for a long time and provides timely warning when the health fluctuates, which is convenient for users to provide dynamic health monitoring management and feedback support.

[0114] 6. Intelligence: the system uses AR augmented reality to present information directly on the real-world view, improving the intelligence of traditional Chinese medicine health monitoring, which is more intuitive and efficient; helps to meet the modern individualized intelligent traditional Chinese medicine health monitoring management needs.

[0115] Further, with reference to Figure 2As shown, the embodiment of the present application further provides an electronic device, which applies the health monitoring system based on pulse signal multi-domain feature extraction and deep learning to assist Chinese medicine health analysis and evaluation; the electronic device can include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and can further include a computer program stored in the memory 11 and executable on the processor 10.

[0116] In some embodiments, the processor 10 can be composed of integrated circuits, for example, can be composed of a single packaged integrated circuit, or can be composed of multiple packaged integrated circuits with the same function or different functions, including one or more combinations of central processing units (CPU), microprocessors, digital processing chips, graphics processors and various control chips. The processor 10 is the control core of the electronic device, which connects various components of the electronic device through various interfaces and lines, and executes programs or modules stored in the memory 11 and calls data stored in the memory 11 to perform various functions and process data of the electronic device.

[0117] It should be understood by those skilled in the art that the embodiments of the present application can be provided as software systems, electronic devices or computer program products, etc. Therefore, the present application can be in the form of a complete software embodiment, a complete hardware embodiment, or an embodiment combining software and hardware aspects.

[0118] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding the components does not exclude the presence of multiple such components. The present application can be implemented by means of hardware comprising several distinct components, and by means of a suitably programmed computer.

[0119] In the specification, the progressive description is adopted, and the same or similar parts between various embodiments can be referred to each other. The unexplained parts of the embodiments of the present application can be obtained from the corresponding product manual or the prior art in the field, which is known in the art, and will not be described in more detail.

[0120] The above describes the embodiments of the present application in detail, and the principles and implementation modes of the present application are described. The above description of the embodiments is only used to help understand the method of the present application and its core idea.

[0121] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and, while certain modifications are discussed, it is desired to be protected in accordance with the spirit and scope of the application. Therefore, the application is not limited to the specific embodiments shown and described, but only by the scope of the appended claims, unless otherwise specified.

Claims

1. A health monitoring system based on multi-domain feature extraction and deep learning of pulse signals, characterized in that: The system includes: a multi-dimensional feature extraction module, a feature association mapping module, a state recognition module and a health information feedback module, wherein: The multi-dimensional feature extraction module uses FMCW radar and sensor equipment to collect and acquire pulse signals, and performs multi-level processing and feature extraction on the acquired pulse signals to extract features related to health status; The feature association mapping module is used to match and associate the pulse signal features of different frequency bands with health status; The state recognition module automatically recognizes and classifies pulse signal characteristics and health status based on a deep learning model; A health information feedback module is used to analyze the user's health status and provide health feedback based on the health status recognition results of the status recognition module, and generate corresponding personalized health guidance information; The multi-dimensional feature extraction module performs multi-level processing and feature extraction on the acquired pulse signal, specifically including: time domain analysis, frequency domain analysis and nonlinear feature analysis, wherein: In time domain analysis, the basic features of the pulse signal are extracted, including the amplitude, peaks, troughs and pulse transmission time of the waveform; In frequency domain analysis, the pulse signal is converted from the time domain to the frequency domain, and its frequency is decomposed to extract the energy distribution of different frequency bands; In nonlinear feature analysis, entropy analysis and Lyapunov exponent method are used to evaluate the dynamic complexity of pulse signals and the stability of organ functions. In the frequency domain analysis, the pulse signal is first converted into frequency domain information through Fourier transform to obtain the energy distribution of the signal at different frequencies; then, it is restored to the time domain signal through inverse Fourier transform, and the analytical signal is generated using Hilbert transform; finally, the analytical signal is combined with the original signal to obtain a signal containing amplitude and phase information. The calculation formula includes: ； Where, Represents a time domain signal The spectrum of the signal after Fourier transformation; represents a frequency variable; Represents the time variable, which is the time position of the signal in the time domain; represents an imaginary unit; represents the complex exponential function; Indicates signal The frequency domain information obtained after Fourier transform is the spectrum representation of the signal. Indicates signal The pulse signal obtained after preprocessing; represents the Hilbert transform to generate the analytical signal, represents the time delay or lag, and in this formula serves as a dummy integration variable in the Hilbert transform process; Represents the analysis signal With signal Composite signal; Represents the instantaneous amplitude information of the signal; Represents the instantaneous phase information of the signal; The feature association mapping module includes: a frequency domain and viscera function association mapping unit, an amplitude and fluctuation amplitude analysis unit and a waveform feature database, wherein: The frequency domain and viscera function association mapping unit is used to associate the pulse signal characteristics of different frequency bands with the functional status of the viscera, and analyze the viscera status according to the fluctuation of specific frequencies; The amplitude and fluctuation amplitude analysis unit is used to analyze the pulse signal envelope after Hilbert transformation to obtain instantaneous amplitude and phase information, judge the strength of the internal organs through the change of amplitude fluctuation, and observe the flow state of qi and blood through the dynamic change of phase; The waveform feature database is used to store pulse diagnosis data and pulse characteristics, providing a pattern reference for judging the functional status of internal organs; the database contains standard pulse signal waveform data under various physical conditions and health conditions, and has a variety of qi, blood and internal organs status patterns.

2. A health monitoring system based on pulse signal multi-domain feature extraction and deep learning according to claim 1, characterized in that: The system also includes: a state tracking and trend analysis module for performing time series analysis based on pulse signal data and drawing a health trend graph that changes over time.

3. The health monitoring system based on multi-domain feature extraction and deep learning of pulse signals according to claim 1, characterized in that: The system also includes: a personalized baseline measurement module for establishing an initial reference baseline for the user's health status, and comparing the initial reference baseline with the user's test results each time to determine abnormal health changes.

4. A health monitoring system based on pulse signal multi-domain feature extraction and deep learning according to claim 2, characterized in that: The system also includes: an AR display module that uses augmented reality technology to present the information collected by the FMCW radar and sensor equipment, the health feedback and personalized health guidance information provided by the health information feedback module, and the health trend chart drawn by the status tracking and trend analysis module on a real-world view.

5. The health monitoring system based on multi-domain feature extraction and deep learning of pulse signals according to claim 1, characterized in that: The operating frequency range of the FMCW radar covers 24 GHz to 77 GHz. The sensor equipment includes: an infrared sensor and a pressure sensor, wherein: the infrared sensor collects optical waveform information of the pulse by detecting changes in light absorption caused by blood flow; the pressure sensor directly captures pressure changes caused by pulse fluctuations through mechanical contact.

6. The health monitoring system based on multi-domain feature extraction and deep learning of pulse signals according to claim 1, characterized in that: The state recognition module uses the ResNet101 deep learning model to achieve automatic recognition and classification of pulse signal characteristics and health status through layer-by-layer convolution.

7. The health monitoring system based on multi-domain feature extraction and deep learning of pulse signals according to claim 1, characterized in that: When analyzing the user's health status, the health information feedback module also conducts a comprehensive evaluation based on the user's pulse information, respiratory rate, and body temperature information. The evaluation method is as follows: ； Where, is the health index, which indicates the health status value after comprehensive evaluation; is a constant term, which represents the basic health index when there is no deviation; The weight coefficient representing the pulse deviation; Indicates the actual pulse value. Indicates the standard value of pulse; represents the weight coefficient of respiratory rate deviation, Indicates the actual respiratory rate, Indicates the standard value of respiratory rate under ideal health conditions; represents the weight coefficient of body temperature deviation, Indicates the current body temperature value. Indicates the standard value of body temperature.

Images