Traditional Chinese medicine qi-blood-body fluid disease identification method and system based on RPPG technology

Abstract

The invention relates to the technical field of biological intelligence, and provides a traditional Chinese medicine qi-blood-body fluid disease recognition method and system based on an RPPG technology, and the method comprises the steps: collecting RGB video data of a patient's face, and extracting an RPPG signal from the RGB video data based on a remote photoplethysmography; noise reduction processing is conducted on the RPPG signals, time domain parameters, frequency domain parameters and nonlinear parameters related to heart rate variability (HRV) are extracted, normalization processing and feature selection are conducted, and feature subsets are obtained; establishing an incidence relation mapping database of the characteristic parameters and the traditional Chinese medicine qi-blood-body fluid state based on the characteristic subsets; establishing a traditional Chinese medicine qi-blood-body fluid state identification model based on the incidence relation mapping database; and obtaining a face video sample of a patient to be identified, extracting a corresponding feature subset to be identified, identifying the feature subset to be identified through the traditional Chinese medicine qi-blood-body fluid state identification model, and outputting an identification result of the traditional Chinese medicine qi-blood-body fluid state. According to the invention, automatic identification, diagnosis and output of traditional Chinese medicine qi-blood-body fluid states are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biological intelligence, and in particular to a traditional Chinese medicine qi-blood-liquor disease recognition method and system based on RPPG technology. BACKGROUND

[0002] Qi, blood and liquor are the basic substances for constituting and maintaining human life activities. Physiologically, qi, blood and liquor are the material basis of the functional activities of zang-fu organs and also the product of the functional activities of zang-fu organs. Pathologically, the pathological changes of zang-fu organs can affect the changes of qi, blood and liquor, and the pathological changes of qi, blood and liquor can also affect certain zang-fu organs. Traditional qi-blood-liquor syndrome differentiation is mainly to analyze the symptoms and signs of patients to determine the gain and loss and running of qi, blood and liquor, so as to determine the related diseases and possible risks. As a non-contact physiological signal detection method, RPPG (Remote Photoplethysmography) technology extracts physiological parameters such as heart rate variability (HRV) from facial video signals collected by a camera. HRV reflects the regulation ability of the autonomic nervous system of the heart, and its time domain, frequency domain and nonlinear parameters can represent the physiological state of the human body from different dimensions, which has inherent relevance with the concept of overall regulation and dynamic balance emphasized in the theory of traditional Chinese medicine qi-blood-liquor.

[0003] In the prior art, traditional Chinese medicine qi-blood-liquor disease recognition mainly adopts direct feature matching and traditional pattern recognition methods to realize the extraction of HRV parameters and the basic health status evaluation function. However, the quantitative correlation between HRV parameters and traditional Chinese medicine qi-blood-liquor syndromes is not considered, it is difficult to organically integrate objective physiological parameters and abstract traditional Chinese medicine theory system, and it is difficult to realize accurate and reliable automatic recognition of traditional Chinese medicine qi-blood-liquor state. SUMMARY

[0004] Therefore, the present application provides a traditional Chinese medicine qi-blood-liquor disease recognition method and system based on RPPG technology, which solves the problem that the quantitative correlation between HRV parameters and traditional Chinese medicine qi-blood-liquor syndromes is not considered in the prior art, it is difficult to organically integrate objective physiological parameters and abstract traditional Chinese medicine theory system, and it is difficult to realize accurate and reliable automatic recognition of traditional Chinese medicine qi-blood-liquor state.

[0005] The technical scheme of the present application is realized as follows: on the one hand, the present application provides a traditional Chinese medicine qi-blood-liquor disease recognition method based on RPPG technology, comprising the following steps:

[0006] acquiring RGB video data of the face of the patient through a camera, and extracting RPPG signals from the RGB video data based on remote photoplethysmography;

[0007] performing noise reduction processing on the RPPG signal, extracting time domain parameters, frequency domain parameters and non-linear parameters related to heart rate variability HRV;

[0008] performing normalization processing and feature selection on the time domain parameters, frequency domain parameters and non-linear parameters to obtain a feature subset with the highest correlation with the state of traditional Chinese medicine qi-blood-liquid;

[0009] Based on the feature subset, a feature parameter and a traditional Chinese medicine qi-blood-liquid state association relationship mapping database is established;

[0010] Based on the association relationship mapping database, a traditional Chinese medicine qi-blood-liquid state identification model is established using a logistic regression algorithm with the feature subset as the input variable;

[0011] Obtain a patient face video sample to be identified, extract a feature subset to be identified corresponding to the patient face video sample to be identified, identify the feature subset to be identified through the traditional Chinese medicine qi-blood-liquid state identification model, and output the identification result of the traditional Chinese medicine qi-blood-liquid state.

[0012] On the basis of the above technical solutions, preferably, based on the feature subset, a feature parameter and a traditional Chinese medicine qi-blood-liquid state association relationship mapping database is established, comprising:

[0013] The feature subset is divided into qi parameter group, blood parameter group and liquid parameter group according to traditional Chinese medicine theory, wherein the time domain parameters are classified into qi parameter group, the frequency domain parameters are classified into blood parameter group, and the non-linear parameters are classified into liquid parameter group;

[0014] Based on the theory of mutual restriction of traditional Chinese medicine qi-blood-liquid, the parameter combination association mode under the combination state of qi deficiency-blood stasis and qi stagnation-liquid stagnation syndrome is calculated, and a multi-syndrome coupling association matrix is established;

[0015] According to the nine constitution classifications of traditional Chinese medicine, a special parameter-syndrome mapping relationship table is established for each constitution type, and a differentiated association relationship mapping database containing constitution weight factors is generated.

[0016] On the basis of the above technical solutions, preferably, based on the theory of mutual restriction of traditional Chinese medicine qi-blood-liquid, the parameter combination association mode under the combination state of qi deficiency-blood stasis and qi stagnation-liquid stagnation syndrome is calculated, and a multi-syndrome coupling association matrix is established, comprising:

[0017] Based on the pathological evolution law of traditional Chinese medicine, a qi deficiency-blood stasis and liquid deficiency-qi deficiency syndrome evolution path model is established, and the parameter change trend coefficient of each stage on the syndrome evolution path is calculated;

[0018] For different syndrome types of the same disease, a differentiation boundary matrix between syndromes is established through differentiated threshold setting of parameter combination.

[0019] For the complex syndrome type of qi-blood syndrome and qi-nutrient syndrome, the contribution weight of each single syndrome type in the complex state is calculated by using a fuzzy membership function, and a multi-dimensional syndrome type weight vector is output.

[0020] On the basis of the above technical scheme, preferably, based on the association relationship mapping database, a traditional Chinese medicine qi-blood-nutrient state identification model is established by using a logistic regression algorithm with the feature subset as an input variable, comprising:

[0021] Based on the association relationship mapping database, a three-level classification structure of qi-blood-nutrient is constructed, wherein the first layer is used to judge the abnormal type of qi, blood, and nutrient system, the second layer is used to judge the specific syndrome type classification, and the third layer is used to output the syndrome type severity level.

[0022] The feature subset is input into a logistic regression classifier, the initial probability value of each syndrome type is calculated, the probability value is fused and corrected in combination with the co-occurrence rule of traditional Chinese medicine syndrome type, and a syndrome type probability distribution vector is generated.

[0023] Based on the constraint relationship of mutual transformation of qi-blood-nutrient in traditional Chinese medicine theory, the parameters of the logistic regression model are optimized, and a traditional Chinese medicine qi-blood-nutrient state identification model is obtained.

[0024] On the basis of the above technical scheme, preferably, the feature subset is input into a logistic regression classifier, the initial probability value of each syndrome type is calculated, the probability value is fused and corrected in combination with the co-occurrence rule of traditional Chinese medicine syndrome type, and a syndrome type probability distribution vector is generated, comprising:

[0025] The feature subset is input into the logistic regression classifier of qi deficiency, qi stagnation, blood deficiency, blood stasis, nutrient deficiency, and nutrient stagnation syndrome type, respectively, and the independent probability value of each syndrome type is calculated.

[0026] Based on the mutual exclusion relationship between syndromes in traditional Chinese medicine theory, the probability values of mutually exclusive syndromes of qi deficiency and qi stagnation, and mutually exclusive syndromes of blood deficiency and blood stasis are normalized and corrected.

[0027] Based on the synergistic relationship between syndromes in traditional Chinese medicine theory, the probability values of qi deficiency-blood stasis synergistic syndrome type combination and qi stagnation-nutrient stagnation synergistic syndrome type combination are subjected to weight enhancement processing, and a syndrome type probability distribution vector subjected to traditional Chinese medicine theory constraint is obtained.

[0028] On the basis of the above technical scheme, preferably, the RGB video data of the patient's face is collected by the camera, and the RPPG signal is extracted from the RGB video data based on the remote photoplethysmography method.

[0029] acquire RGB video data of a patient's face region through a camera at a preset frame rate, and pre-process the RGB video data to obtain standard RGB video data;

[0030] extract an RPPG signal reflecting periodic changes in blood volume from changes in a green channel component in the standard RGB video data based on a principle of a photoplethysmogram.

[0031] On the basis of the above technical solutions, preferably, the RPPG signal is subjected to noise reduction processing, and time-domain parameters, frequency-domain parameters, and nonlinear parameters related to heart rate variability (HRV) are extracted, including:

[0032] The RPPG signal is subjected to filtering and noise reduction and artifact removal processing to obtain a heart rate variability analysis signal;

[0033] Based on the heart rate variability analysis signal, time-domain parameters, frequency-domain parameters, and nonlinear parameters related to heart rate variability (HRV) are extracted, the time-domain parameters including MEAN, SDNN, RMSSD, NN50, PNN50, and CV, the frequency-domain parameters including VLF, LF, HF, TP, and LF / HF, and the nonlinear parameters including SD1, SD2, SD1 / SD2, and SampEn.

[0034] On the basis of the above technical solutions, preferably, the time-domain parameters, frequency-domain parameters, and nonlinear parameters are subjected to normalization processing and feature selection to obtain a feature subset with the highest correlation with the state of traditional Chinese medicine qi-blood-humor, including:

[0035] The time-domain parameters, frequency-domain parameters, and nonlinear parameters are subjected to data preprocessing and normalization processing to eliminate differences in dimensions and numerical ranges among different parameters, and a standardized parameter set after normalization processing is obtained;

[0036] Based on traditional Chinese medicine qi-blood-humor state annotation data, a feature selection algorithm is used to screen the standardized parameter set to obtain a feature subset with the highest correlation with the state of traditional Chinese medicine qi-blood-humor.

[0037] On the basis of the above technical solutions, preferably, the patient face video sample to be identified is acquired, a feature subset to be identified corresponding to the patient face video sample to be identified is extracted, the feature subset to be identified is identified through the traditional Chinese medicine qi-blood-humor state identification model, and an identification result of the state of traditional Chinese medicine qi-blood-humor is output, including:

[0038] The patient face video sample to be identified is subjected to segmentation processing in a sliding time window manner, HRV feature parameters corresponding to each time window are extracted, and the HRV feature parameters of multiple time windows are fused to obtain a time sequence fusion feature vector;

[0039] input the time sequence fusion feature vector into the traditional Chinese medicine qi-blood-juice state identification model to obtain an identification result corresponding to the patient to be identified, the identification result including a syndrome type classification result, a syndrome type probability distribution, an identification confidence, and a time change trend.

[0040] In another aspect, the application also provides a traditional Chinese medicine qi-blood-juice disease identification system based on RPPG technology, the system comprising:

[0041] a data acquisition module configured to acquire RGB video data of a patient's face through a camera and extract RPPG signals from the RGB video data based on remote photoplethysmography;

[0042] a parameter extraction module configured to perform noise reduction processing on the RPPG signals and extract time domain parameters, frequency domain parameters, and nonlinear parameters related to heart rate variability (HRV);

[0043] a feature selection module configured to perform normalization processing and feature selection on the time domain parameters, frequency domain parameters, and nonlinear parameters to obtain a feature subset with the highest correlation with traditional Chinese medicine qi-blood-juice state;

[0044] a mapping establishment module configured to establish a mapping database of the correlation between feature parameters and traditional Chinese medicine qi-blood-juice state based on the feature subset;

[0045] a model establishment module configured to establish a traditional Chinese medicine qi-blood-juice state identification model based on the mapping database of the correlation and using the feature subset as input variables by using a logistic regression algorithm;

[0046] a disease identification module configured to acquire a patient's face video sample to be identified, extract a feature subset to be identified corresponding to the patient's face video sample to be identified, identify the feature subset to be identified by using the traditional Chinese medicine qi-blood-juice state identification model, and output an identification result of traditional Chinese medicine qi-blood-juice state.

[0047] The traditional Chinese medicine qi-blood-juice disease identification method and system based on RPPG technology have the following beneficial effects compared with the prior art:

[0048] (1) By integrating RPPG signal extraction, HRV multi-dimensional parameter analysis, feature selection optimization, and correlation mapping modeling, an identification model is dynamically established based on a logistic regression algorithm, a non-contact video detection architecture is used, a quantitative correlation mechanism between feature parameters and traditional Chinese medicine syndromes is introduced, time domain, frequency domain, and nonlinear parameters are mapped with traditional Chinese medicine qi-blood-juice state, objective physiological parameters are highly integrated with traditional Chinese medicine theory system, and automatic identification and diagnosis and output of traditional Chinese medicine qi-blood-juice state are realized.

[0049] (2) By mapping the HRV characteristic parameters according to the grouping of traditional Chinese medicine theory, a multi- syndrome type coupling correlation matrix is constructed, and a constitution differentiation mapping relationship is established. Based on the mutual restraint theory of traditional Chinese medicine of qi, blood and body fluid, the syndrome type combination correlation mode is dynamically calculated. By using the nine constitution classification framework of traditional Chinese medicine, the syndrome type evolution path modeling and composite syndrome type weight distribution mechanism are introduced. The objective physiological parameters and abstract traditional Chinese medicine theory are deeply integrated to ensure that the quantitative correlation between the characteristic parameters and the traditional Chinese medicine syndrome types is established in accordance with the logical relationship of traditional Chinese medicine theory. The individualized, differentiated and accurate mapping of the state of traditional Chinese medicine of qi, blood and body fluid and the objective syndrome differentiation under the guidance of theory are realized.

[0050] (3) By constructing a three-level classification structure of qi, blood and body fluid, implementing syndrome type probability fusion calculation and traditional Chinese medicine theory constraint optimization, dynamically calculating the initial probability of each syndrome type based on the logistic regression algorithm and fusion correction, using a hierarchical judgment framework, introducing syndrome type mutual exclusion correction and synergy enhancement processing mechanism, ensuring that the model output is highly consistent with the principles of traditional Chinese medicine syndrome differentiation and treatment, and realizing the hierarchical identification, probability distribution calculation and result output of the state of traditional Chinese medicine of qi, blood and body fluid. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0052] Figure 1 A flow chart of a traditional Chinese medicine qi, blood and body fluid disease recognition method based on RPPG technology. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be described in detail below. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0054] Please refer to Figure 1 The present application provides a traditional Chinese medicine qi, blood and body fluid disease recognition method based on RPPG technology, which comprises the following steps:

[0055] The RGB video data of the patient's face is collected by the camera, and the RPPG signal is extracted from the RGB video data based on the remote photoplethysmography method;

[0056] The RPPG signal is denoised to extract time domain parameters, frequency domain parameters and nonlinear parameters related to heart rate variability (HRV);

[0057] The time domain parameters, frequency domain parameters and nonlinear parameters are normalized and selected to obtain a feature subset most relevant to the state of traditional Chinese medicine (TCM) Qi, blood and body fluid;

[0058] Based on the feature subset, a mapping database of feature parameters and the state of TCM Qi, blood and body fluid is established;

[0059] Based on the mapping database of the association, a TCM Qi, blood and body fluid state recognition model is established using a logistic regression algorithm with the feature subset as the input variable;

[0060] A patient face video sample to be identified is obtained, and a feature subset to be identified corresponding to the patient face video sample to be identified is extracted. The feature subset to be identified is identified by the TCM Qi, blood and body fluid state recognition model, and an identification result of the state of TCM Qi, blood and body fluid is output.

[0061] Specifically, the present embodiment integrates RPPG signal extraction, HRV multi-dimensional parameter analysis, feature selection optimization and correlation mapping modeling, dynamically establishes a recognition model based on a logistic regression algorithm, uses a non-contact video detection architecture, introduces a quantitative correlation mechanism between feature parameters and TCM syndromes, maps time domain, frequency domain and nonlinear parameters to the state of TCM Qi, blood and body fluid, ensures the high integration of objective physiological parameters and traditional TCM theoretical system, and realizes the automatic recognition and diagnosis of the state of TCM Qi, blood and body fluid and output.

[0062] The RGB video data of the patient's face is collected by the camera, and the RPPG signal is extracted from the RGB video data based on remote photoplethysmography (RPPG), which comprises:

[0063] The RGB video data of the patient's face region is collected by the camera at a preset frame rate, and the RGB video data is preprocessed to obtain standard RGB video data.

[0064] In a specific embodiment, the RGB video data of the patient's face region is collected by the camera at a preset frame rate, and the RGB video data is preprocessed to obtain standard RGB video data, which comprises:

[0065] The frame rate of the camera is set to 25-30 fps, and the resolution is not less than 640x480 pixels;

[0066] Haar cascade classifier or deep learning algorithm is used to detect and locate the face region in the video frame;

[0067] extract a region of interest (ROI) rich in capillaries such as forehead, both cheeks, etc. in the face region;

[0068] perform illumination compensation and motion artifact correction on the region of interest.

[0069] extract a RPPG signal reflecting periodic changes in blood volume from changes in the green channel component in the standard RGB video data based on the principle of photoplethysmography.

[0070] In a specific embodiment, the extraction of the RPPG signal reflecting periodic changes in blood volume from changes in the green channel component in the standard RGB video data based on the principle of photoplethysmography comprises:

[0071] separate the pre-processed RGB video data into red (R), green (G) and blue (B) channels;

[0072] extract the average pixel intensity time series of the green (G) channel in the region of interest as the original pulse signal;

[0073] perform filtering processing on the original pulse signal using a band-pass filter with a frequency range of 0.5-4 Hz, corresponding to a heart rate range of 30-240 bpm;

[0074] perform noise reduction and enhancement processing on the filtered signal using independent component analysis (ICA) or principal component analysis (PCA) algorithm to obtain a high-quality RPPG signal.

[0075] Specifically, the embodiment realizes stable extraction of high-quality RPPG signals by integrating camera acquisition parameter optimization, face detection and positioning, region of interest extraction, and illumination and motion correction, dynamically extracting green channel blood volume change signals based on the principle of photoplethysmography, using a multi-algorithm fusion processing architecture, introducing ICA or PCA noise reduction and enhancement, and a multi-stage filtering mechanism, ensuring a reliable conversion relationship between the original video data and physiological signals, and realizing precise acquisition of non-contact physiological signals, enhanced anti-interference capability, and significant improvement in signal quality.

[0076] The noise reduction processing on the RPPG signal to extract time-domain, frequency-domain and non-linear parameters related to heart rate variability (HRV) comprises:

[0077] perform filtering noise reduction and artifact removal processing on the RPPG signal to obtain a high-quality heart rate variability analysis signal.

[0078] In a specific embodiment, the filtering noise reduction and artifact removal processing on the RPPG signal to obtain a high-quality heart rate variability analysis signal comprises:

[0079] a high-pass filter is used to remove baseline drift components with a frequency lower than 0.05 Hz in the RPPG signal;

[0080] a band-pass filter is used to retain the heart rate variability effective signal in the frequency band of 0.05-0.5 Hz, and filter out high-frequency noise and power frequency interference;

[0081] an adaptive filtering algorithm or wavelet transform method is used to detect and remove the artifact signal caused by head movement;

[0082] a moving average filter or Savitzky-Golay filter is used for signal smoothing processing to maintain the time domain characteristics of the signal.

[0083] Based on the heart rate variability analysis signal, time domain parameters, frequency domain parameters and nonlinear parameters related to heart rate variability HRV are extracted, the time domain parameters include MEAN, SDNN, RMSSD, NN50, PNN50, CV, the frequency domain parameters include VLF, LF, HF, TP, LF / HF, and the nonlinear parameters include SD1, SD2, SD1 / SD2, SampEn.

[0084] In a specific embodiment, based on the heart rate variability analysis signal, time domain parameters, frequency domain parameters and nonlinear parameters related to heart rate variability HRV are extracted, including:

[0085] The peak value detection algorithm is used to identify the heart beat peak value in the heart rate variability analysis signal, and the R-R interval sequence between adjacent heart beats is calculated;

[0086] Based on the R-R interval sequence, the time domain parameters MEAN, SDNN, RMSSD, NN50, PNN50 and CV are calculated, wherein MEAN is the average value of R-R interval, SDNN is the standard deviation of R-R interval, RMSSD is the root mean square of adjacent R-R interval difference, NN50 is the number of adjacent R-R interval difference greater than 50 ms, PNN50 is the percentage of NN50 in the total number of heart beats, and CV is the coefficient of variation;

[0087] The R-R interval sequence is subjected to fast Fourier transform FFT, and the frequency domain parameters VLF, LF, HF, TP and LF / HF are calculated, wherein VLF is the power of very low frequency 0.003-0.04 Hz, LF is the power of low frequency 0.04-0.15 Hz, HF is the power of high frequency 0.15-0.4 Hz, TP is the total power, and LF / HF is the ratio of low frequency to high frequency power;

[0088] Based on the Poincare scatter plot analysis method, the nonlinear parameters SD1, SD2 and SD1 / SD2 are calculated, wherein SD1 is the standard deviation of the short axis of the ellipse, SD2 is the standard deviation of the long axis of the ellipse, and SD1 / SD2 is the ratio of the short axis to the long axis, and the sample entropy algorithm is used to calculate the SampEn parameter to reflect the complexity of the signal.

[0089] Specifically, the embodiment integrates multi-stage filtering denoising, artifact removal processing, peak detection algorithm and multi-dimensional parameter extraction, dynamically calculates time domain, frequency domain and nonlinear parameters based on heart rate variability analysis theory, adopts adaptive filtering and wavelet transform architecture, introduces R-R interval sequence analysis, fast Fourier transform and Poincare scatter plot analysis mechanism, stably extracts high-quality HRV signals, calculates multi-dimensional physiological parameters, ensures the establishment of a reliable quantitative relationship between RPPG signals and heart rate variability parameters, and realizes high-precision analysis of physiological signals, enhanced anti-interference capability and comprehensive representation of multi-dimensional parameters.

[0090] The time domain parameters, frequency domain parameters and nonlinear parameters are normalized and feature selected to obtain a feature subset with the highest correlation with the state of traditional Chinese medicine Qi, blood and body fluid, including:

[0091] The time domain parameters, frequency domain parameters and nonlinear parameters are preprocessed and normalized to eliminate the dimensional and numerical range differences between different parameters, and a standardized parameter set after normalization is obtained.

[0092] In a specific embodiment, the time domain parameters, frequency domain parameters and nonlinear parameters are preprocessed and normalized to eliminate the dimensional and numerical range differences between different parameters, and a standardized parameter set after normalization is obtained, including:

[0093] Anomalies and outliers in the time domain parameters, frequency domain parameters and nonlinear parameters are detected and removed using statistical methods;

[0094] Each parameter is converted to a standard normal distribution or a specified distribution form using Z-score standardization or other standardization methods;

[0095] The parameters are mapped to a preset numerical interval using minimum-maximum normalization or other normalization methods;

[0096] To address the problem of unbalanced sample quantity for different traditional Chinese medicine syndromes, oversampling or undersampling methods are used to balance the sample quantity of each category, and a balanced standardized parameter set is output.

[0097] Based on the traditional Chinese medicine Qi, blood and body fluid state annotation data, a feature selection algorithm is used to screen the standardized parameter set to obtain a feature subset with the highest correlation with the traditional Chinese medicine Qi, blood and body fluid state.

[0098] In a specific embodiment, based on the traditional Chinese medicine Qi, blood and body fluid state annotation data, a feature selection algorithm is used to screen the standardized parameter set to obtain a feature subset with the highest correlation with the traditional Chinese medicine Qi, blood and body fluid state, including:

[0099] The correlation between each HRV parameter and the traditional Chinese medicine Qi, blood and body fluid state is calculated using correlation analysis methods;

[0100] The importance score of each parameter to the classification of the state of traditional Chinese medicine Qi-blood-Body fluid is evaluated by information theory or statistical methods;

[0101] The optimal feature subset is selected by using recursive feature elimination, forward selection or backward elimination, etc. Feature selection algorithm combined with cross-validation technology;

[0102] The stability and generalization ability of the selected feature subset are verified by cross-validation method to ensure the reliability of the feature subset.

[0103] Specifically, the present embodiment integrates the abnormal value detection process, multiple standardization methods optimization, sample balancing technology and feature selection algorithm, dynamically evaluates the importance of parameters based on correlation analysis and information theory method, uses recursive feature elimination and cross-validation architecture, introduces stability verification and generalization ability evaluation mechanism, standardizes the HRV multi-dimensional parameters, and selects the optimal feature subset, ensures the establishment of unified quantitative standard between different dimension parameters, realizes the high-quality preprocessing, dimension optimization and significant improvement of the correlation with traditional Chinese medicine syndrome type of feature parameters.

[0104] The mapping database of the correlation between the feature parameters and the state of traditional Chinese medicine Qi-blood-Body fluid is established based on the feature subset, including:

[0105] The feature subset is divided into Qi parameter group, blood parameter group and Body fluid parameter group according to traditional Chinese medicine theory, wherein the time domain parameters are classified into Qi parameter group, the frequency domain parameters are classified into blood parameter group, and the nonlinear parameters are classified into Body fluid parameter group;

[0106] Based on the theory of mutual restriction of traditional Chinese medicine Qi-blood-Body fluid, the parameter combination correlation mode under the combined state of Qi deficiency-blood stasis and Qi stagnation-Body fluid stagnation syndrome type is calculated, and a multi-syndrome type coupling correlation matrix is established.

[0107] According to the nine constitution classifications of traditional Chinese medicine, a dedicated parameter-syndrome mapping table is established for each constitution type, and a differentiated correlation mapping database containing constitution weight factors is generated.

[0108] The mapping database of the correlation between the feature parameters and the state of traditional Chinese medicine Qi-blood-Body fluid is established based on the feature subset, including:

[0109] Based on the pathological evolution law of traditional Chinese medicine, an evolution path model of Qi deficiency-blood stasis and Body fluid deficiency-Qi deficiency syndrome type is established, and the parameter change trend coefficient of each stage on the evolution path is calculated.

[0110] For different syndrome types of the same disease, a differentiation boundary matrix between syndromes is established by setting different threshold values for parameter combinations.

[0111] For the complex syndrome type of qi-blood syndrome and qi-fluid syndrome, the fuzzy membership function is used to calculate the contribution weight of each single syndrome type in the complex state, and output the multi-dimensional syndrome weight vector.

[0112] In a specific embodiment, the calculation formula of the constitution weight factor is:

[0113]

[0114] wherein w ij is the weight factor of the ith constitution type to the jth parameter group; α i is the basic weight coefficient of the ith constitution type, α k is the basic weight coefficient of the kth constitution type, i, k ∈ {1, 2, …, 9}; β j is the importance coefficient of the jth parameter group (meteorological, hematological, and fluidic), β l is the importance coefficient of the lth parameter group, j, l ∈ {1, 2, 3}; μ ij is the association distance of the ith constitution type and the jth parameter group, μ kl is the association distance of the kth constitution type and the lth parameter group; exp(·) is the exponential function, used to realize the nonlinear weight distribution.

[0115] It should be noted that the traditional weight distribution method usually adopts linear weighting or simple normalization. In this embodiment, the fusion calculation of three dimensions of constitution type basic weight (α i ), parameter group importance coefficient (β j ), and association distance (μ ij ) is introduced, and the nonlinear weight attenuation is realized through the exponential function exp(-μ ij ), so that the weight distribution is more in line with the theoretical requirements of individualized syndrome differentiation of traditional Chinese medicine.

[0116] This embodiment realizes the personalized parameter weight self-adaptive distribution based on nine constitution types, improves the syndrome type identification accuracy of patients with different constitutions, avoids the weight distribution defects of traditional methods, and makes the relevance of HRV parameters and traditional Chinese medicine theory more closely.

[0117] In a specific embodiment, the calculation formula of the multi-syndrome type coupling association matrix is:

[0118]

[0119] wherein R mn is the coupling association strength of the mth syndrome type and the nth syndrome type; δ mn is the Kronecker function, when m = n, δ mn is 1, otherwise it is 0; tanh(·) is the hyperbolic tangent activation function; ξq is the weight parameter of the qth path of qi-blood-body fluid evolution; φ mnq is the correlation coefficient of syndrome type m and n on the qth path of qi-blood-body fluid evolution; σ(·) is a Sigmoid activation function; ζ r is the strength parameter of the rth inter-syndrome constraint relationship; ψ mnr is the influence factor of syndrome type m and n under the rth inter-syndrome constraint relationship; Q is the total number of qi-blood-body fluid evolution paths; R is the total number of inter-syndrome constraint relationships.

[0120] The embodiment changes the traditional independent classification mode, adopts a Kronecker function δ mn The embodiment distinguishes between self-association and mutual association of syndrome types, processes qi-blood-body fluid evolution paths by using a hyperbolic tangent activation function tanh(·) for self-association, processes inter-syndrome constraint relationships by using a Sigmoid activation function for mutual association, and constructs a complex nonlinear coupling relationship model between syndrome types.

[0121] The embodiment establishes a mathematical model in accordance with the TCM theories of “syndrome types being compatible” and “mutual transformation”, effectively captures the synergistic, mutually exclusive, and evolution relationships between syndrome types, improves the recognition accuracy of complex syndrome types and syndrome transformation, and enhances the TCM theory interpretability of the recognition results.

[0122] Specifically, the embodiment maps HRV feature parameters according to TCM theories, constructs a multi-syndrome coupling correlation matrix, and establishes a body constitution differentiation mapping relationship, dynamically calculates a syndrome combination correlation pattern based on the TCM theory of qi-blood-body fluid mutual restraint, adopts a TCM nine-constitution classification framework, introduces a syndrome evolution path modeling and complex syndrome weight distribution mechanism, realizes the deep integration of objective physiological parameters and abstract TCM theories, ensures the establishment of a quantitative correlation relationship between feature parameters and TCM syndromes in accordance with TCM theoretical logic, and realizes individualized, differentiated TCM qi-blood-body fluid state precise mapping and objective syndrome differentiation under theoretical guidance.

[0123] Based on the correlation relationship mapping database, a TCM qi-blood-body fluid state recognition model is established by using a logistic regression algorithm with the feature subset as an input variable, including:

[0124] Based on the correlation relationship mapping database, a qi-blood-body fluid three-level classification structure is constructed, wherein the first level is used to determine the abnormal type of qi, blood, and body fluid systems, the second level is used to determine the specific syndrome classification, and the third level is used to output the syndrome severity level;

[0125] The feature subset is input into a logistic regression classifier, the initial probability value of each syndrome type is calculated, the probability value is fused and corrected in combination with the TCM syndrome co-occurrence rule, and a syndrome probability distribution vector is generated;

[0126] Based on the constraint relationship of mutual transformation of Qi, blood and body fluid in traditional Chinese medicine theory, the parameters of the logistic regression model are optimized, and the identification model of Qi, blood and body fluid state in traditional Chinese medicine is output, which conforms to the logic of traditional Chinese medicine theory.

[0127] The feature subset is input into a logistic regression classifier, the initial probability value of each syndrome type is calculated, the probability value is fused and corrected combined with the co-occurrence rule of traditional Chinese medicine syndrome type, and a syndrome probability distribution vector is generated.

[0128] The feature subset is input into the logistic regression classifier of each single syndrome type such as Qi deficiency, Qi stagnation, blood deficiency, blood stasis, body fluid deficiency and body fluid stagnation, and the independent probability value of each syndrome type is calculated.

[0129] Based on the mutual exclusion relationship between syndromes in traditional Chinese medicine theory, the probability values of mutually exclusive syndromes of Qi deficiency and Qi stagnation, and blood deficiency and blood stasis are normalized and corrected.

[0130] Based on the synergistic relationship between syndromes in traditional Chinese medicine theory, the probability values of Qi deficiency-blood stasis synergistic syndrome combination and Qi stagnation-body fluid stagnation synergistic syndrome combination are enhanced by weight, and the syndrome probability distribution vector constrained by traditional Chinese medicine theory is obtained.

[0131] In a specific embodiment, the initial probability value calculation formula of each syndrome type of the Qi-blood-body fluid three-level classification structure is:

[0132]

[0133] Wherein, is the initial probability of the kth syndrome of the fth layer of the tth abnormal system, f∈{1,2,3} respectively corresponds to three layers of abnormal type judgment, specific syndrome classification and severity level; exp(·) is the exponential function; is the weight parameter of the kth syndrome of the fth layer of the tth type to the hth feature, is the weight parameter of the gth syndrome of the fth layer of the tth type to the hth feature; x h is the hth feature value in the input feature subset; is the bias parameter of the kth syndrome of the fth layer of the tth type, is the bias parameter of the gth syndrome of the fth layer of the tth type; N (f) is the total number of syndromes in the fth layer; H is the dimension of the feature subset.

[0134] This embodiment extends the traditional single-layer softmax classifier to a three-level structure, each layer undertakes different classification tasks (abnormal type judgment, specific syndrome classification, severity level), and realizes layer-by-layer refined probability calculation through hierarchical weight parameters and bias , which avoids the problems of high dimension and feature confusion in single-layer classification.

[0135] The embodiment realizes gradual syndrome type identification from coarse granularity to fine granularity, significantly reduces classification complexity, improves accuracy and stability of each level classification, and makes the identification process more consistent with the logical thinking mode of TCM clinical diagnosis.

[0136] In a specific embodiment, the calculation formula of the syndrome type probability distribution vector is:

[0137]

[0138] wherein, is the final modified probability of the kth syndrome type; is the second layer initial probability of the kth syndrome type, is the second layer initial probability of the cth syndrome type, is the second layer initial probability of the sth syndrome type, is the second layer initial probability of the uth syndrome type; S k is the syndrome type set having a synergistic relationship with the syndrome type k, S c is the syndrome type set having a synergistic relationship with the syndrome type c; U c is the syndrome type set having a mutual exclusion relationship with the syndrome type c; γ s is the enhancement coefficient of the synergistic syndrome type s; θ u is the inhibition coefficient of the mutually exclusive syndrome type u; T is the total number of syndrome types.

[0139] The embodiment introduces a double modification mechanism of the synergistic relationship set S k and the mutual exclusion relationship set U c on the basis of the initial probability, dynamically adjusts the probability distribution of each syndrome type through the multiplication operation of the synergistic enhancement coefficient γ s and the mutual exclusion inhibition coefficient θ u , and embodies the interaction law of the syndrome types of TCM “Xiang Xu Xiang Shizhe” and “Xiang Wei Xiang Sha”.

[0140] The embodiment effectively solves the problem that the traditional independent probability output ignores the mutual influence between the syndrome types, improves the identification rationality and clinical credibility under the condition of multiple coexisting syndrome types, and ensures that the output result conforms to the overall concept of TCM syndrome differentiation and treatment.

[0141] Specifically, the embodiment realizes the construction of a syndrome type identification model consistent with the logical theory of TCM by constructing a three-level classification structure of Qi, blood and body fluid, implementing syndrome type probability fusion calculation and TCM theory constraint optimization, dynamically calculating the initial probability of each syndrome type based on the logistic regression algorithm and fusing and modifying it, adopting a hierarchical judgment architecture, and introducing a mutual exclusion correction and synergistic enhancement processing mechanism, so as to ensure that the model output is highly consistent with the principle of TCM syndrome differentiation and treatment, realize the hierarchical identification of the state of Qi, blood and body fluid of TCM, the probability distribution calculation under the constraint of the theory, and the logical syndrome differentiation result output.

[0142] The acquisition of the patient face video sample to be identified, the extraction of the corresponding to-be-identified feature subset of the patient face video sample to be identified, the identification of the to-be-identified feature subset by the traditional Chinese medicine qi-blood-body fluid state identification model, and the output of the identification result of the traditional Chinese medicine qi-blood-body fluid state include:

[0143] The to-be-identified patient face video sample is segmented by using a sliding time window mode, the HRV feature parameters corresponding to each time window are extracted, and the HRV feature parameters of multiple time windows are fused to obtain a time sequence fusion feature vector.

[0144] In an embodiment, the to-be-identified patient face video sample is segmented by using a sliding time window mode, the HRV feature parameters corresponding to each time window are extracted, and the HRV feature parameters of multiple time windows are fused to obtain a time sequence fusion feature vector, including:

[0145] The to-be-identified patient face video sample is segmented according to a preset time length, each segment has a duration of 30-60 seconds, and the overlap degree between adjacent segments is 50%;

[0146] The corresponding to-be-identified feature subset of each video segment is extracted according to the processing procedure of the foregoing steps;

[0147] According to the distance from the current time, the feature subsets of the video segments are assigned time sequence attenuation weights, the weight of a recent segment is higher, and the weight of a long-term segment is lower;

[0148] The weighted feature subsets of the video segments are linearly weighted and fused to generate a comprehensive time sequence fusion feature vector.

[0149] The time sequence fusion feature vector is input into the traditional Chinese medicine qi-blood-body fluid state identification model to obtain the identification result corresponding to the to-be-identified patient, and the identification result includes a syndrome type classification result, a syndrome type probability distribution, an identification confidence, and a time change trend.

[0150] In an embodiment, the time sequence fusion feature vector is input into the traditional Chinese medicine qi-blood-body fluid state identification model to obtain the identification result corresponding to the to-be-identified patient, and the identification result includes a syndrome type classification result, a syndrome type probability distribution, an identification confidence, and a time change trend, including:

[0151] Based on the time sequence fusion feature vector, the probability values of each syndrome type are calculated by the traditional Chinese medicine qi-blood-body fluid state identification model, and the syndrome type with the highest probability value is selected as the main syndrome type classification result.

[0152] Based on the difference between the highest probability value and the second highest probability value and the similarity between the feature vector and the training sample, the confidence score of the current identification result is calculated.

[0153] The current recognition result is compared with the historical recognition result, a time change gradient of a syndrome type probability is calculated, and a syndrome type evolution trend index is outputted.

[0154] The syndrome type classification result, the probability distribution, the confidence score and the trend index are combined, and a structured multi-dimensional recognition result data is outputted.

[0155] Specifically, the embodiment integrates sliding time window segmentation processing, multi-window HRV parameter fusion, time sequence attenuation weight distribution and multi-dimensional recognition result output, dynamically calculates a time sequence fusion feature vector based on a time sequence analysis algorithm, adopts a video segmentation processing architecture, introduces a historical data comparison and confidence evaluation mechanism, so that the recognition result is highly consistent with the time sequence change rule of the real state of the patient, and dynamic tracking identification, trend prediction analysis and confidence quantitative evaluation of the TCM syndrome type are realized.

[0156] The application also provides a TCM qi-blood-liquor disease recognition system based on the RPPG technology.

[0157] The data acquisition module is configured to acquire RGB video data of a patient's face through a camera and extract RPPG signals from the RGB video data based on the remote photoplethysmography method.

[0158] The parameter extraction module is configured to perform noise reduction processing on the RPPG signals and extract time domain parameters, frequency domain parameters and non-linear parameters related to heart rate variability (HRV).

[0159] The feature selection module is configured to perform normalization processing and feature selection on the time domain parameters, frequency domain parameters and non-linear parameters to obtain a feature subset with the highest correlation with the TCM qi-blood-liquor state.

[0160] The mapping establishment module is configured to establish a correlation mapping database between feature parameters and the TCM qi-blood-liquor state based on the feature subset.

[0161] The model establishment module is configured to establish a TCM qi-blood-liquor state recognition model based on the correlation mapping database and using the feature subset as input variables by adopting a logistic regression algorithm.

[0162] The disease recognition module is configured to acquire a patient face video sample to be identified, extract a feature subset to be identified corresponding to the patient face video sample to be identified, identify the feature subset to be identified by using the TCM qi-blood-liquor state recognition model, and output a recognition result of the TCM qi-blood-liquor state.

[0163] Specifically, the TCM qi-blood-liquor disease recognition system based on the RPPG technology integrates the non-contact video processing of the data acquisition module, the multi-dimensional feature analysis of the parameter extraction module, the optimized screening of the feature selection module, the correlation relationship construction of the mapping establishment module, the intelligent learning of the model establishment module, and the automatic output of the disease recognition module, dynamically constructs a recognition model based on the RPPG signal extraction and HRV parameter analysis fusion technology, adopts a modular system architecture, introduces a logistic regression algorithm optimization and a TCM theory constraint mechanism, realizes the full-process automatic processing from video acquisition to syndrome type recognition, ensures the collaborative cooperation and smooth data flow among the modules of the system, and realizes the intelligent recognition and standardized diagnosis process of the TCM qi-blood-liquor disease.

[0164] The above merely describes preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A traditional Chinese medicine qi-blood-liquor disease identification method based on RPPG technology, characterized in that, The method comprises the following steps: acquiring RGB video data of a patient's face through a camera and extracting a RPPG signal from the RGB video data based on remote photoplethysmography; performing noise reduction processing on the RPPG signal, extracting time-domain parameters, frequency-domain parameters and nonlinear parameters related to heart rate variability (HRV), and performing normalization processing and feature selection on the time-domain parameters, frequency-domain parameters and nonlinear parameters to obtain a feature subset with the highest correlation with the state of traditional Chinese medicine (TCM) qi-blood-liquid; based on the feature subset, establishing a mapping database of the correlation between feature parameters and the state of TCM qi-blood-liquid; based on the mapping database of the correlation, establishing a TCM qi-blood-liquid state identification model using a logistic regression algorithm with the feature subset as input variables; acquiring a patient face video sample to be identified, extracting a feature subset to be identified corresponding to the patient face video sample to be identified, identifying the feature subset to be identified through the TCM qi-blood-liquid state identification model, and outputting an identification result of the state of TCM qi-blood-liquid. The method comprises the following steps:

2. The method according to claim 1, wherein the method is characterized in that, dividing the feature subset into qi parameter groups, blood parameter groups and liquid parameter groups according to TCM theory, wherein the time-domain parameters are classified into qi parameter groups, the frequency-domain parameters are classified into blood parameter groups, and the nonlinear parameters are classified into liquid parameter groups; based on the theory of mutual restraint of TCM qi-blood-liquid, calculating the parameter combination correlation mode in the combined state of qi deficiency-blood stasis and qi stagnation-liquid stagnation syndrome types, and establishing a multi-syndrome type coupling correlation matrix; for each constitution type, establishing a dedicated parameter-syndrome type mapping table to generate a differentiated correlation mapping database containing constitution weight factors. The method comprises the following steps:

3. The method according to claim 2, wherein the method is characterized in that, based on the theory of mutual restraint of TCM qi-blood-liquid, calculating the parameter combination correlation mode in the combined state of qi deficiency-blood stasis and qi stagnation-liquid stagnation syndrome types, and establishing a multi-syndrome type coupling correlation matrix, which comprises: based on the pathological evolution law of TCM, establishing a qi deficiency-blood stasis and liquid deficiency-qi deficiency syndrome evolution path model, and calculating the parameter change trend coefficient at each stage of the syndrome evolution path; for different syndrome types of the same disease, a differentiated threshold of parameter combination is set to establish a differentiation boundary matrix between different syndrome types; for qi-blood disease and qi-liquid disease complex syndrome types, a fuzzy membership function is used to calculate the contribution weight of each single syndrome type in the complex state, and a multi-dimensional syndrome weight vector is output.

4. The method according to claim 1, wherein the method is characterized in that, The method comprises the following steps: based on the mapping database of the correlation, a three-level classification structure of qi-blood-liquid is constructed, wherein the first layer is used to judge the abnormal type of qi, blood and liquid systems, the second layer is used to judge the specific syndrome type classification, and the third layer is used to output the severity level of the syndrome type; the feature subset is input into a logistic regression classifier to calculate the initial probability value of each syndrome type, and the probability value is fused and corrected in combination with the co-occurrence law of TCM syndrome types to generate a syndrome type probability distribution vector. Based on the constraint relationship of mutual transformation of Qi, blood and body fluid in traditional Chinese medicine theory, the parameters of the logistic regression model are optimized, and the identification model of Qi, blood and body fluid state in traditional Chinese medicine is obtained.

5. The traditional Chinese medicine blood and body fluid disease identification method based on RPPG technology according to claim 4, characterized in that, The feature subset is input into a logistic regression classifier, and the initial probability value of each syndrome is calculated. The probability value is fused and corrected combined with the co-occurrence rule of traditional Chinese medicine syndrome to generate a syndrome probability distribution vector, including: The feature subset is input into a logistic regression classifier for each syndrome, and the independent probability value of each syndrome is calculated. Based on the mutual exclusion relationship between syndromes in traditional Chinese medicine theory, the probability values of mutually exclusive syndromes of Qi deficiency and Qi stagnation, and blood deficiency and blood stasis are normalized and corrected. Based on the synergistic relationship between syndromes in traditional Chinese medicine theory, the probability values of Qi deficiency-blood stasis synergistic syndrome combination and Qi stagnation-turbid fluid stagnation synergistic syndrome combination are enhanced by weight to obtain a syndrome probability distribution vector constrained by traditional Chinese medicine theory.

6. The traditional Chinese medicine blood and body fluid disease recognition method based on RPPG technology according to claim 1, characterized in that, The RGB video data of the patient's face is collected by the camera, and the RPPG signal is extracted from the RGB video data based on remote photoplethysmography. RGB video data of the patient's face region is collected by the camera at a preset frame rate, and the RGB video data is preprocessed to obtain standard RGB video data. Based on the principle of photoplethysmography, the RPPG signal reflecting the periodic change of blood volume is extracted from the green channel component change in the standard RGB video data.

7. The traditional Chinese medicine blood and body fluid disease recognition method based on RPPG technology according to claim 1, characterized in that, The RPPG signal is denoised, and time domain parameters, frequency domain parameters and non-linear parameters related to heart rate variability (HRV) are extracted. The RPPG signal is filtered and denoised, and artifacts are removed to obtain heart rate variability analysis signals. Based on the heart rate variability analysis signal, time domain parameters, frequency domain parameters and non-linear parameters related to heart rate variability (HRV) are extracted. The time domain parameters include MEAN, SDNN, RMSSD, NN50, PNN50 and CV. The frequency domain parameters include VLF, LF, HF, TP and LF / HF. The non-linear parameters include SD1, SD2, SD1 / SD2 and SampEn.

8. The traditional Chinese medicine blood and body fluid disease recognition method based on RPPG technology according to claim 1, characterized in that, The time domain parameters, frequency domain parameters and non-linear parameters are normalized and selected to obtain a feature subset with the highest correlation with the state of Qi, blood and body fluid in traditional Chinese medicine. The time domain parameters, frequency domain parameters and non-linear parameters are preprocessed and normalized to eliminate the differences in dimensions and numerical ranges between different parameters to obtain a set of standardized parameters after normalization. Based on the labeled data of Qi, blood and body fluid state in traditional Chinese medicine, a feature selection algorithm is used to filter the standardized parameter set to obtain a feature subset with the highest correlation with the state of Qi, blood and body fluid in traditional Chinese medicine.

9. The traditional Chinese medicine blood and body fluid disease recognition method based on RPPG technology according to claim 1, characterized in that, The face video sample of the patient to be identified is obtained, the corresponding feature subset of the patient to be identified is extracted, the feature subset is identified by the identification model of Qi, blood and body fluid state in traditional Chinese medicine, and the identification result of the state of Qi, blood and body fluid in traditional Chinese medicine is output. The face video sample of the to-be-identified patient is segmented by using a sliding time window mode, HRV feature parameters corresponding to each time window are extracted, and the HRV feature parameters of multiple time windows are fused to obtain a time sequence fusion feature vector; The time sequence fusion feature vector is input into the TCM qi-blood-liquid state identification model to obtain an identification result corresponding to the to-be-identified patient, and the identification result includes a syndrome type classification result, a syndrome type probability distribution, an identification confidence, and a time variation trend.

10. A traditional Chinese medicine blood and fluid disease identification system based on RPPG technology, used for executing a traditional Chinese medicine blood and fluid disease identification method based on RPPG technology according to any one of claims 1-9, characterized in that, The system comprises: A data acquisition module configured to acquire RGB video data of a patient's face by using a camera and extract a RPPG signal from the RGB video data based on a remote photoplethysmography method; A parameter extraction module configured to perform noise reduction processing on the RPPG signal and extract time domain parameters, frequency domain parameters, and nonlinear parameters related to heart rate variability (HRV); A feature selection module configured to perform normalization processing and feature selection on the time domain parameters, frequency domain parameters, and nonlinear parameters to obtain a feature subset most relevant to TCM qi-blood-liquid state; A mapping establishment module configured to establish a mapping database of the correlation between feature parameters and TCM qi-blood-liquid state based on the feature subset; A model establishment module configured to establish a TCM qi-blood-liquid state identification model based on the mapping database of the correlation and using the feature subset as input variables by using a logistic regression algorithm; A disease identification module configured to acquire a face video sample of a to-be-identified patient, extract a to-be-identified feature subset corresponding to the face video sample of the to-be-identified patient, identify the to-be-identified feature subset by using the TCM qi-blood-liquid state identification model, and output an identification result of TCM qi-blood-liquid state.

Images