Method and system for identifying yin and yang syndromes in traditional Chinese medicine based on fusion of bioelectricity signal characteristics

Abstract

The invention relates to the technical field of traditional Chinese medicine yin-yang syndrome identification, and provides a traditional Chinese medicine yin-yang syndrome identification method and system based on fusion of bioelectricity signal characteristics, and the method comprises the steps: collecting a bioelectricity signal of a detected object, and constructing a feature vector; constructing an error function; constructing a neural network model by taking the feature vector as input; training a neural network model by using the error function to obtain a first yin-yang syndrome identification model; constructing a support vector machine model and a three-layer syndrome decision tree model, and performing series integration on the support vector machine model and the three-layer syndrome decision tree model to obtain a second yin-yang syndrome identification model; performing fusion processing on the first yin-yang syndrome identification model and the second yin-yang syndrome identification model to obtain a fusion model; and inputting the feature vector into the fusion model for processing, and outputting an identification result. According to the invention, deep fusion of traditional Chinese medicine theory constraint and a modern machine learning technology is realized, and objectivity of yin and yang syndrome identification is improved.

Description

Technical Field

[0001] This invention relates to the field of TCM Yin-Yang syndrome identification technology, and in particular to a TCM Yin-Yang syndrome identification method and system based on the fusion of bioelectrical signal characteristics. Background Technology

[0002] The identification of Yin-Yang syndromes in Traditional Chinese Medicine (TCM) is a core component of the traditional TCM diagnostic system. It guides clinical diagnosis and treatment by recognizing the balance of Yin and Yang in the human body. Yin-Yang theory, as the foundation of TCM theory, posits that human health is maintained by the balance and coordination of Yin and Yang, while the occurrence and development of disease manifest as different syndrome types of Yin-Yang imbalance. In modern TCM clinical practice, accurate identification of Yin-Yang syndromes is crucial for developing personalized treatment plans and is widely used in key medical areas such as chronic disease management, sub-health assessment, and constitution identification. However, traditional Yin-Yang syndrome identification methods, in long-term clinical application, are affected by factors such as differences in physician experience, subjective judgment bias, and the complexity of syndrome manifestations, making it difficult to guarantee the consistency and accuracy of identification results.

[0003] In existing technologies, TCM syndrome identification based on bioelectric signals mainly employs traditional feature extraction and basic machine learning classification methods, achieving preliminary syndrome identification functions. However, existing methods do not adequately consider the intrinsic physiological correlation mechanism between TCM Yin-Yang theory and bioelectric signal features, making it difficult to organically integrate objective bioelectric field physiological constraints with actual Yin-Yang syndrome theoretical features, thus failing to achieve high-precision automatic identification of TCM Yin-Yang syndrome states. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on the fusion of bioelectric signal features. This solves the problem that the existing technology does not adequately consider the intrinsic physiological correlation mechanism between Yin-Yang theory in Traditional Chinese Medicine and bioelectric signal features, making it difficult to organically integrate the objective physiological constraints of bioelectric field with the actual Yin-Yang syndrome theoretical features, thus making it impossible to achieve high-precision automatic identification of Yin-Yang syndrome states.

[0005] The technical solution of this invention is implemented as follows: On the one hand, this invention provides a method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on the fusion of bioelectrical signal features, comprising the following steps: Bioelectric signals of the tested object are acquired by remote photoplethysmography, and a multidimensional bioelectric signal feature vector is constructed based on the bioelectric signals. Based on the theory of the unity of opposites in Yin-Yang syndromes and the theory of hierarchical weighting of syndromes, a multi-label error function constrained by the unity of opposites in Yin-Yang is constructed. Using the multidimensional bioelectric signal feature vector as input, a Yin-Yang balance perception BP-MLL neural network model is constructed; The Yin-Yang balance perception BP-MLL neural network model is trained using the Yin-Yang opposition and unity constraint multi-label error function to obtain the first Yin-Yang syndrome identification model. A support vector machine model and a three-layer syndrome decision tree model are constructed, and the support vector machine model and the three-layer syndrome decision tree model are cascaded and integrated to obtain a second yin-yang syndrome identification model. The first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model are fused together using a confidence-weighted fusion algorithm and a dynamic rule base update mechanism to obtain a TCM theory-guided decision fusion model. The multidimensional bioelectric signal feature vector is input into the TCM theory-guided decision fusion model for processing, and the multi-label yin-yang syndrome identification result is output.

[0006] Based on the above technical solutions, preferably, the construction of a multi-label error function constrained by the unity of opposites in Yin-Yang syndromes and the hierarchical theory of syndrome weights includes: By setting a penalty function for the combination of opposing Yin and Yang syndromes, a Yin-Yang opposition constraint term is constructed to constrain the simultaneous prediction results of opposing syndromes. By setting differentiated weight coefficients based on the clinical importance level of TCM syndromes, a syndrome weight constraint term is constructed to differentiate the prediction error of primary syndrome, secondary syndrome, and concurrent syndrome. The Yin-Yang opposition constraint term and the syndrome weight constraint term are combined with the traditional multi-label loss term to obtain the Yin-Yang opposition unity constraint multi-label error function.

[0007] Based on the above technical solutions, preferably, the penalty function for the Yin-Yang syndrome opposition combination includes: Establish a set of Yin syndrome clusters and a set of Yang syndrome clusters, and set mutual exclusion constraint rules between Yin syndrome clusters and Yang syndrome clusters. When the output of the neural network includes the predicted values ​​of both Yin syndrome clusters and Yang syndrome clusters, input the predicted values ​​of Yin syndrome clusters and Yang syndrome clusters into a penalty function for calculation to obtain the corresponding penalty weight values. Differentiate the penalty weight values ​​from the predicted values ​​within the same type of syndrome cluster, and output the constraint term values ​​after processing by the penalty function.

[0008] Based on the above technical solutions, preferably, the step of constructing a Yin-Yang balance perception BP-MLL neural network model using the multidimensional bioelectric signal feature vector as input includes: The Yin-Yang balance perception BP-MLL neural network model includes an input layer, a hidden layer, a Yin-Yang balance perception layer, and an output layer. A yin-yang balance perception layer is constructed, which includes a yin-yang syndrome perception mechanism. The yin-yang balance perception layer is set between the hidden layer and the output layer and is used to receive the feature representation of the hidden layer as input. A yin-yang mutual exclusion gating unit is set in the yin-yang balance perception layer. The yin-yang mutual exclusion gating unit includes a yin syndrome gating channel and a yang syndrome gating channel. The yin syndrome gating channel and the yang syndrome gating channel selectively transmit yin syndrome-related features and yang syndrome-related features, respectively. In the Yin-Yang balance perception layer, an HRV feature-guided adaptive activation function is set. The parameter configuration of the HRV feature-guided adaptive activation function is adjusted according to the type of input heart rate variability parameters. The gated features are input into the HRV feature-guided adaptive activation function for nonlinear transformation, and the processing result of the Yin-Yang balance perception layer is output.

[0009] Based on the above technical solutions, preferably, the male-female mutual exclusion gating unit includes: Establish a Yin syndrome feature index table and a Yang syndrome feature index table, and map the input feature vector to the Yin syndrome feature subset and the Yang syndrome feature subset respectively according to the Yin syndrome feature index table and the Yang syndrome feature index table; Set a gate weight matrix for negative syndromes and a gate weight matrix for positive syndromes, and perform matrix operations on the gate weight matrix for negative syndromes and the gate weight matrix for positive syndromes with the feature subsets for negative syndromes and the feature subsets for positive syndromes, respectively; The sigmoid gating function was used to calculate the gating values ​​for yin syndrome and yang syndrome respectively. The yin syndrome gate value and the yin syndrome feature subset are multiplied element-wise, and the yang syndrome gate value and the yang syndrome feature subset are multiplied element-wise to obtain the yin and yang syndrome feature representations after mutual exclusion gate filtering.

[0010] Based on the above technical solutions, preferably, the step of training the Yin-Yang balance perception BP-MLL neural network model using the Yin-Yang opposition and unity constraint multi-label error function to obtain the first Yin-Yang syndrome identification model includes: A Yin-Yang constraint gradient training mechanism is established, which decomposes the gradient information of the Yin-Yang opposition and unity constraint multi-label error function into the Yin-Yang opposition constraint gradient and the syndrome weight constraint gradient. Set a Yin-Yang balance adjustment factor, and dynamically adjust the value of the Yin-Yang balance adjustment factor according to the ratio of the prediction error of Yin syndrome to the prediction error of Yang syndrome in the current training round. The Yin-Yang balance adjustment factor is used to weight the Yin-Yang opposition constraint gradient, and the weighted Yin-Yang opposition constraint gradient is fused with the syndrome weight constraint gradient to obtain the Yin-Yang perception training gradient. The parameters of the Yin-Yang balance perception BP-MLL neural network model are updated by backpropagation using the Yin-Yang perception training gradient.

[0011] Based on the above technical solutions, preferably, the yin-yang constraint gradient training mechanism includes: Calculate the partial derivatives of the multi-label error function constrained by the Yin-Yang opposition and unity with respect to the network parameters, and calculate the partial derivatives separately according to the Yin-Yang opposition constraint term and the syndrome weight constraint term; Establish a Yin-Yang gradient allocation matrix and allocate the gradient components of the Yin-Yang opposition constraint terms to the Yin syndrome gating channel parameters and Yang syndrome gating channel parameters of the Yin-Yang balance perception layer, respectively. The gradient components of the syndrome weight constraint term are allocated to the corresponding network layer parameters according to the weight ratio of primary syndrome, secondary syndrome, and concurrent syndrome. Set a gradient clipping threshold to restrict the numerical range of each gradient component after allocation, and output the training gradient combination after yin-yang constraint processing.

[0012] Based on the above technical solutions, preferably, the construction of the support vector machine model and the three-layer syndrome decision tree model, and the cascading integration of the support vector machine model and the three-layer syndrome decision tree model to obtain the second yin-yang syndrome identification model, includes: A support vector machine model including constraints from traditional Chinese medicine theory is constructed. These constraints are based on the theory of Yin-Yang and Five Elements in traditional Chinese medicine to set the mutual generation and restraint relationships between syndromes. The decision boundary of the support vector machine is adjusted with theoretical guidance, and the classification results of the support vector machine are output. A three-layer syndrome decision tree model is constructed, comprising a basic judgment layer for Yin deficiency and Yang deficiency, a subdivision layer for Zang-Fu syndrome, and a syndrome severity assessment layer. The support vector machine classification results are used as inputs to the three-layer syndrome decision tree model for concatenation processing to obtain a second Yin-Yang syndrome identification model.

[0013] Based on the above technical solutions, preferably, the step of inputting the multidimensional bioelectric signal feature vector into the TCM theory-guided decision fusion model for processing and outputting multi-label Yin-Yang syndrome identification results includes: The multidimensional bioelectric signal feature vector is standardized and preprocessed. The preprocessed feature vector is then input into the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model in the TCM theory-guided decision fusion model for parallel inference calculation. The fusion prediction probability vector is obtained through a confidence-weighted fusion algorithm. The fused prediction probability vector is binarized using a multi-label threshold discrimination mechanism. Differential discrimination thresholds are set according to the yin-yang syndrome type. Multi-label yin-yang syndrome identification results are generated by combining confidence information. The multi-label yin-yang syndrome identification results include syndrome type, prediction probability, and confidence.

[0014] On the other hand, the present invention also provides a TCM Yin-Yang syndrome identification system based on the fusion of bioelectrical signal features, the system comprising: The feature extraction module is used to acquire the bioelectrical signals of the test object through remote photoplethysmography and construct a multidimensional bioelectrical signal feature vector based on the bioelectrical signals. The Yin-Yang opposition and unity constraint module is used to construct a multi-label error function for Yin-Yang opposition and unity constraint based on the Yin-Yang syndrome opposition and unity theory and the syndrome weight hierarchy theory. The Yin-Yang balance perception module is used to construct a Yin-Yang balance perception BP-MLL neural network model using the multidimensional bioelectric signal feature vector as input. The constraint-guided model training module is used to train the yin-yang balance perception BP-MLL neural network model using the yin-yang opposition and unity constraint multi-label error function to obtain the first yin-yang syndrome identification model. The machine learning model integration module is used to construct a support vector machine model and a three-layer syndrome decision tree model. The support vector machine model and the three-layer syndrome decision tree model are cascaded and integrated to obtain a second yin-yang syndrome identification model. The fusion decision module is used to fuse the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model through a confidence-weighted fusion algorithm and a dynamic rule base update mechanism to obtain a TCM theory-guided decision fusion model. The multi-label syndrome identification output module is used to input the multi-dimensional bioelectric signal feature vector into the TCM theory-guided decision fusion model for processing and output the multi-label yin-yang syndrome identification result.

[0015] The present invention provides a method and system for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on the fusion of bioelectrical signal features, which has the following advantages over the prior art: (1) Objective data acquisition of bioelectric signals is carried out by remote photoplethysmography pulse wave, constrained neural network modeling is carried out based on the Yin-Yang theory of traditional Chinese medicine, and a multi-model confidence weighted fusion intelligent decision-making mechanism is adopted to eliminate the subjectivity and experience dependence in traditional Chinese medicine syndrome identification, realize the deep integration of traditional Chinese medicine theory constraints and modern machine learning technology, and improve the objectivity of Yin-Yang syndrome identification. (2) By establishing the Yin-Yang opposition constraint term and syndrome weight constraint term based on TCM theory, the traditional machine learning multi-label loss function is transformed into a special error function that integrates TCM Yin-Yang theory, thus achieving a training objective that meets both mathematical optimization requirements and follows the laws of TCM theory, and improving the theoretical consistency of Yin-Yang syndrome identification. (3) By constructing a series integration architecture of a support vector machine model and a three-layer syndrome decision tree model that integrates the constraints of traditional Chinese medicine theory, the traditional machine learning method is deeply integrated with the five elements theory and the Yin-Yang deficiency-excess classification system of traditional Chinese medicine. This achieves progressive intelligent identification from basic Yin-Yang classification to organ subdivision and degree assessment, thereby improving the accuracy of syndrome identification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a traditional Chinese medicine Yin-Yang syndrome identification method based on the fusion of bioelectrical signal features, according to the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 This invention provides a method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on the fusion of bioelectrical signal features, comprising the following steps: Bioelectric signals of the tested object are acquired by remote photoplethysmography, and a multidimensional bioelectric signal feature vector is constructed based on the bioelectric signals. Based on the theory of the unity of opposites in Yin-Yang syndromes and the theory of hierarchical weighting of syndromes, a multi-label error function constrained by the unity of opposites in Yin-Yang is constructed. Using the multidimensional bioelectric signal feature vector as input, a Yin-Yang balance perception BP-MLL neural network model is constructed; The Yin-Yang balance perception BP-MLL neural network model is trained using the Yin-Yang opposition and unity constraint multi-label error function to obtain the first Yin-Yang syndrome identification model. A support vector machine model and a three-layer syndrome decision tree model are constructed, and the support vector machine model and the three-layer syndrome decision tree model are cascaded and integrated to obtain a second yin-yang syndrome identification model. The first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model are fused together using a confidence-weighted fusion algorithm and a dynamic rule base update mechanism to obtain a TCM theory-guided decision fusion model. The multidimensional bioelectric signal feature vector is input into the TCM theory-guided decision fusion model for processing, and the multi-label yin-yang syndrome identification result is output.

[0020] Specifically, this embodiment uses remote photoplethysmography to objectively collect bioelectrical signals, models constrained neural networks based on the Yin-Yang theory of Traditional Chinese Medicine, and adopts an intelligent decision-making mechanism that uses multi-model confidence weighted fusion. This eliminates the subjectivity and experience dependence in traditional Chinese medicine syndrome identification, achieves a deep integration of the constraints of traditional Chinese medicine theory and modern machine learning technology, and improves the objectivity of Yin-Yang syndrome identification.

[0021] The method involves acquiring bioelectrical signals from the subject using remote photoplethysmography (PPG), and constructing a multidimensional bioelectrical signal feature vector based on these signals, including: Remote photoplethysmography (PPG) is used to perform non-contact optical detection on the facial area of ​​the subject, acquiring raw bioelectrical signals reflecting cardiovascular activity.

[0022] In one specific embodiment, a light signal of a specific wavelength is emitted to the area of ​​facial color change of the subject, and the reflected light signal modulated by facial vascular pulsation is received. Through photoelectric conversion and signal amplification processing, a raw bioelectric signal containing heart rate rhythm information is obtained. The raw bioelectric signal is used to reflect the activity state of the sympathetic and parasympathetic nervous systems.

[0023] The heart rate variability time-domain parameters, frequency-domain parameters, and nonlinear parameters related to Yin-Yang syndrome are extracted from the original bioelectric signal. After standardizing the heart rate variability time-domain parameters, frequency-domain parameters, and nonlinear parameters, they are combined to obtain a multidimensional bioelectric signal feature vector.

[0024] In one specific embodiment, RMSSD, pNN50, and SDNN are extracted from the original bioelectric signal as time-domain parameters, LF, HF, and LF / HF ratio are extracted as frequency-domain parameters, and approximate entropy and sample entropy are extracted as nonlinear parameters. Based on the sympathetic hyperactivity characteristics corresponding to Yin syndrome and the parasympathetic hyperactivity characteristics corresponding to Yang syndrome, the parameters are weighted according to Yin-Yang syndrome. The weighted parameters are combined to form a multidimensional bioelectric signal feature vector.

[0025] Specifically, this embodiment achieves non-contact detection of facial vascular pulsation through remote photoplethysmography, transforming the traditional contact-based bioelectrical signal acquisition method into a remote detection technology based on optical principles. This solves the problems of subject discomfort, signal interference, and detection limitations associated with traditional contact methods such as electrocardiograms and pulse sensors. It achieves objective, convenient, and non-invasive extraction of heart rate variability parameters and Yin-Yang syndrome-guided feature weight allocation, improving the comfort of bioelectrical signal acquisition and the relevance of feature extraction to traditional Chinese medicine theory.

[0026] The above-mentioned theory of the unity of opposites in Yin-Yang syndromes and the hierarchical theory of syndrome weights constructs a multi-label error function constrained by the unity of opposites in Yin-Yang syndromes, including: By setting a penalty function for the combination of opposing Yin and Yang syndromes, a Yin-Yang opposition constraint term is constructed to constrain the simultaneous prediction results of opposing syndromes. By setting differentiated weight coefficients based on the clinical importance level of TCM syndromes, a syndrome weight constraint term is constructed to differentiate the prediction error of primary syndrome, secondary syndrome, and concurrent syndrome. The Yin-Yang opposition constraint term and the syndrome weight constraint term are combined with the traditional multi-label loss term to obtain the Yin-Yang opposition unity constraint multi-label error function.

[0027] The penalty function for the opposing combination of Yin and Yang syndromes includes: Establish a set of Yin syndrome clusters and a set of Yang syndrome clusters, and set mutual exclusion constraint rules between Yin syndrome clusters and Yang syndrome clusters. When the output of the neural network includes the predicted values ​​of both Yin syndrome clusters and Yang syndrome clusters, input the predicted values ​​of Yin syndrome clusters and Yang syndrome clusters into a penalty function for calculation to obtain the corresponding penalty weight values. Differentiate the penalty weight values ​​from the predicted values ​​within the same type of syndrome cluster, and output the constraint term values ​​after processing by the penalty function.

[0028] In one specific embodiment, the formula for calculating the yin-yang opposition constraint term is: ; in, The loss value is the constraint term representing the Yin-Yang opposition. The total number of training samples; For the set index of negative syndromes, Index for the set of positive syndromes; For the first One sample of negative syndrome The predicted probability, For the first One sample showed positive symptoms. The predicted probability; The weighting coefficient for the Yin-Yang opposition constraint; The activation threshold is the mutual exclusion of yin and yang. The attenuation factor for the distance of the syndrome; Yin syndrome Harmony Yang Syndrome The function of the strength of opposition between them.

[0029] The formula for calculating the syndrome weight constraint term is: ; in, The loss value is the weighted loss term for the symptoms. This represents the total number of syndrome categories; For symptoms The clinical importance weighting function; For symptoms Hierarchical weighting (primary certificate / secondary certificate / concurrent certificate); For the first Symptoms of a sample The true label; For the sample number Symptoms of a sample The predicted probability; The error amplification index; For the first Symptoms of a sample The moderating factor of syndrome severity.

[0030] Specifically, this embodiment transforms the traditional machine learning multi-label loss function into a dedicated error function that integrates the Yin-Yang theory of traditional Chinese medicine by establishing a Yin-Yang opposition constraint term and a syndrome weight constraint term based on traditional Chinese medicine theory. This achieves a training objective that meets both mathematical optimization requirements and follows the laws of traditional Chinese medicine theory, thereby improving the theoretical consistency of Yin-Yang syndrome identification.

[0031] The step of constructing a Yin-Yang balance perception BP-MLL neural network model using the multidimensional bioelectric signal feature vector as input includes: The Yin-Yang balance perception BP-MLL neural network model includes an input layer, a hidden layer, a Yin-Yang balance perception layer, and an output layer. A yin-yang balance perception layer is constructed, which includes a yin-yang syndrome perception mechanism. The yin-yang balance perception layer is set between the hidden layer and the output layer and is used to receive the feature representation of the hidden layer as input. A yin-yang mutual exclusion gating unit is set in the yin-yang balance perception layer. The yin-yang mutual exclusion gating unit includes a yin syndrome gating channel and a yang syndrome gating channel. The yin syndrome gating channel and the yang syndrome gating channel selectively transmit yin syndrome-related features and yang syndrome-related features, respectively. In the Yin-Yang balance perception layer, an HRV feature-guided adaptive activation function is set. The parameter configuration of the HRV feature-guided adaptive activation function is adjusted according to the type of input heart rate variability parameters. The gated features are input into the HRV feature-guided adaptive activation function for nonlinear transformation, and the processing result of the Yin-Yang balance perception layer is output.

[0032] The mutual exclusion gate unit includes: Establish a Yin syndrome feature index table and a Yang syndrome feature index table, and map the input feature vector to the Yin syndrome feature subset and the Yang syndrome feature subset respectively according to the Yin syndrome feature index table and the Yang syndrome feature index table; Set a gate weight matrix for negative syndromes and a gate weight matrix for positive syndromes, and perform matrix operations on the gate weight matrix for negative syndromes and the gate weight matrix for positive syndromes with the feature subsets for negative syndromes and the feature subsets for positive syndromes, respectively; The sigmoid gating function was used to calculate the gating values ​​for yin syndrome and yang syndrome respectively. The yin syndrome gate value and the yin syndrome feature subset are multiplied element-wise, and the yang syndrome gate value and the yang syndrome feature subset are multiplied element-wise to obtain the yin and yang syndrome feature representations after mutual exclusion gate filtering.

[0033] Specifically, this embodiment transforms the traditional neural network backpropagation algorithm into a dedicated training method that integrates the Yin-Yang theory of Traditional Chinese Medicine by establishing a Yin-Yang constraint gradient training mechanism and a dynamic balance adjustment factor. This solves the problems of existing deep learning models ignoring the constraints of Traditional Chinese Medicine theory, failing to achieve balanced learning of Yin-Yang syndromes, and having uneven gradient distribution. It realizes an intelligent training process that satisfies both the convergence requirements of mathematical optimization and the Yin-Yang balance concept of Traditional Chinese Medicine, thereby improving the neural network model's ability to learn Yin-Yang syndrome features and its prediction accuracy.

[0034] The process of training the Yin-Yang balance-aware BP-MLL neural network model using the Yin-Yang opposition and unity constraint multi-label error function to obtain the first Yin-Yang syndrome identification model includes: A Yin-Yang constraint gradient training mechanism is established, which decomposes the gradient information of the Yin-Yang opposition and unity constraint multi-label error function into the Yin-Yang opposition constraint gradient and the syndrome weight constraint gradient. Set a Yin-Yang balance adjustment factor, and dynamically adjust the value of the Yin-Yang balance adjustment factor according to the ratio of the prediction error of Yin syndrome to the prediction error of Yang syndrome in the current training round. The Yin-Yang balance adjustment factor is used to weight the Yin-Yang opposition constraint gradient, and the weighted Yin-Yang opposition constraint gradient is fused with the syndrome weight constraint gradient to obtain the Yin-Yang perception training gradient. The parameters of the Yin-Yang balance perception BP-MLL neural network model are updated by backpropagation using the Yin-Yang perception training gradient.

[0035] The yin-yang constraint gradient training mechanism includes: Calculate the partial derivatives of the multi-label error function constrained by the Yin-Yang opposition and unity with respect to the network parameters, and calculate the partial derivatives separately according to the Yin-Yang opposition constraint term and the syndrome weight constraint term; Establish a Yin-Yang gradient allocation matrix and allocate the gradient components of the Yin-Yang opposition constraint terms to the Yin syndrome gating channel parameters and Yang syndrome gating channel parameters of the Yin-Yang balance perception layer, respectively. The gradient components of the syndrome weight constraint term are allocated to the corresponding network layer parameters according to the weight ratio of primary syndrome, secondary syndrome, and concurrent syndrome. Set a gradient clipping threshold to restrict the numerical range of each gradient component after allocation, and output the training gradient combination after yin-yang constraint processing.

[0036] Specifically, this embodiment constructs a cascaded integrated architecture that combines a support vector machine model constrained by traditional Chinese medicine theory with a three-layer syndrome decision tree model. This deeply integrates traditional machine learning methods with the five elements theory and the yin-yang classification system of traditional Chinese medicine, achieving progressive intelligent identification from basic yin-yang classification to organ subdivision and then to degree assessment, thereby improving the accuracy of syndrome identification.

[0037] The construction of the support vector machine model and the three-layer syndrome decision tree model, and the cascading integration of the support vector machine model and the three-layer syndrome decision tree model to obtain the second yin-yang syndrome identification model, includes: A support vector machine model including constraints from traditional Chinese medicine theory is constructed. These constraints are based on the theory of Yin-Yang and Five Elements in traditional Chinese medicine, which sets the mutual generation and restraint relationships between syndromes. The decision boundary of the support vector machine is adjusted with theoretical guidance, and the classification results of the support vector machine are output.

[0038] In one specific embodiment, a matrix of mutual generation and restraint relationships of TCM syndromes is established, quantifying the syndrome relationships of the five internal organs and six bowels in traditional Chinese medicine theory into numerical constraints. A syndrome relationship penalty term is added to the objective function of the support vector machine. The syndrome relationship penalty term applies penalty weights to the predicted combinations that violate traditional Chinese medicine theory based on the mutual generation and restraint relationship matrix. The support vector machine optimization problem with constraints from traditional Chinese medicine theory is solved by using the Lagrange multiplier method. The decision hyperplane parameters that conform to the constraints of traditional Chinese medicine theory are obtained, and the support vector machine classification results corrected by traditional Chinese medicine theory are output.

[0039] A three-layer syndrome decision tree model is constructed, comprising a basic judgment layer for Yin deficiency and Yang deficiency, a subdivision layer for Zang-Fu syndrome, and a syndrome severity assessment layer. The support vector machine classification results are used as inputs to the three-layer syndrome decision tree model for concatenation processing to obtain a second Yin-Yang syndrome identification model.

[0040] In one specific embodiment, the objective function of the support vector machine model constrained by traditional Chinese medicine theory is calculated as follows: ; in, The overall objective function of the support vector machine is constrained by traditional Chinese medicine theory. These are the support vector machine weight vectors; For bias terms; For the first Slack variables for each sample; The penalty parameters are those for traditional SVM. The total number of training samples; The weighting coefficients are constrained by traditional Chinese medicine theory. The number of negative symptoms, The number of positive symptoms. For symptoms Symptoms The elements of the weight matrix representing the mutual generation and restraint relationships of the five elements; and Symptoms Symptoms The corresponding weight sub-vectors; For the first Feature vectors of training samples; This is a transpose operation.

[0041] The output calculation formula of the second Yin-Yang syndrome identification model is: ; in, For the cascade ensemble model to input samples The final predicted probability; It is the sigmoid activation function; For the first Linear combination weights of a layered decision tree; For the first Layer decision trees, given the historical information of the previous layer Predicted probability under given conditions The output of the support vector machine is used as the initial condition; The weights of the support vector machine in the ensemble; The interaction coefficients of the three-layer decision tree; Corresponding to the basic judgment level of Yin deficiency and Yang deficiency, Corresponding to the subdivision of visceral syndromes, Corresponding to the syndrome severity assessment layer.

[0042] In the basic judgment layer for Yin deficiency and Yang deficiency, a Yin deficiency discrimination threshold and a Yang deficiency discrimination threshold are set. Basic Yin-Yang classification is performed based on the comparison results between the feature values ​​output by the support vector machine and the Yin deficiency discrimination threshold and the Yang deficiency discrimination threshold. In the subdivision layer of visceral syndromes, five visceral syndrome branches of heart, liver, spleen, lung and kidney are established. Based on the output of the basic judgment layer of yin deficiency and yang deficiency, the feature vectors are assigned to the corresponding visceral syndrome branches for subdivision and identification. In the syndrome severity assessment layer, three severity assessment nodes are set up: mild, moderate and severe. The severity of syndromes of each organ is quantitatively assessed by dividing the feature value interval, and the decision tree classification results containing syndrome type and severity information are output.

[0043] Specifically, this embodiment integrates deep learning models and traditional machine learning models by constructing a confidence-weighted fusion algorithm and a dynamic rule base update mechanism. Combined with the dynamic update mechanism of the TCM theory rule base, it solves the problems of limitations of single models, static model fusion weights, and lack of TCM theory post-processing correction. It realizes adaptive weight allocation based on prediction uncertainty and dynamic rule updates driven by theoretical consistency, thereby improving the prediction accuracy and theoretical compliance of the fusion model.

[0044] The process of fusing the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model using a confidence-weighted fusion algorithm and a dynamic rule base update mechanism to obtain a TCM theory-guided decision-making fusion model includes: The prediction confidence weights of the first and second yin-yang syndrome identification models are calculated using a confidence-weighted fusion algorithm. Dynamic weight coefficients are assigned based on the prediction accuracy of each model for different syndrome types. The outputs of the two models are then weighted and fused to obtain a fusion model.

[0045] In one specific embodiment, a syndrome classification confidence assessment matrix is ​​established, and the prediction probability and prediction uncertainty of the first yin-yang syndrome identification model and the second yin-yang syndrome identification model for each yin-yang syndrome type are calculated respectively. Set up a confidence quantification function based on the variance of the prediction probability to convert the prediction uncertainty of each model into a confidence weight value; The confidence weights of the two models are normalized using the softmax function to obtain normalized weight coefficients. The normalized weight coefficients are weighted and summed with the prediction vectors of the corresponding models to output the fused multi-label prediction results.

[0046] A set of TCM theory-guided rules is established through a dynamic rule base update mechanism. The set of TCM theory-guided rules is updated based on the consistency evaluation between the prediction results of the fusion model and the TCM theory rules. The fusion results are then post-processed and corrected based on the updated rule set to obtain the TCM theory-guided decision fusion model.

[0047] In one specific embodiment, a traditional Chinese medicine theoretical rule base is established, including the rules of mutual exclusion of Yin and Yang, the rules of mutual generation and restraint of the five internal organs, and the rules of progressive syndrome severity. A rule matching degree calculation module is set up to calculate the matching degree between the prediction results of the fusion model and the rule conditions in the TCM theory rule base. When the matching degree is lower than the set threshold, the rule update mechanism is activated to generate new rule entries based on the feature vector of the current predicted sample and the training data label results. The newly generated rule entries are added to the TCM theory rule base. The weights of historical rules are adjusted using a rule weight decay algorithm, and an updated dynamic rule base is output for prediction result correction.

[0048] Specifically, this embodiment establishes a differentiated discrimination mechanism between the threshold sets for Yin syndromes and the threshold sets for Yang syndromes. It binarizes the fused prediction probability vector and generates structured multi-label syndrome identification results. This solves the problems of traditional methods, such as lack of targeted threshold settings, inability to quantify prediction confidence, and insufficient intuitiveness of output results. It realizes adaptive threshold discrimination based on TCM theory and completes a comprehensive output including syndrome type, prediction probability, and confidence information, thereby improving the practicality of syndrome identification results.

[0049] The process of inputting the multidimensional bioelectric signal feature vector into the TCM theory-guided decision fusion model for processing and outputting multi-label Yin-Yang syndrome identification results includes: The multidimensional bioelectric signal feature vector is standardized and preprocessed. The preprocessed feature vector is then input into the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model in the TCM theory-guided decision fusion model for parallel inference calculation. The fusion prediction probability vector is obtained through a confidence-weighted fusion algorithm.

[0050] In one specific embodiment, the Z-score normalization method is used to normalize the feature vector of the multidimensional bioelectric signal, and the mean and standard deviation of the feature vector are calculated to adjust the numerical range. The standardized feature vectors are simultaneously input into the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model to obtain the neural network prediction vector and the decision tree prediction vector, respectively. Based on the confidence weight coefficients calculated during the pre-training phase, the neural network prediction vector and the decision tree prediction vector are weighted and summed to output a fused prediction probability vector that has been corrected for confidence.

[0051] The fused prediction probability vector is binarized using a multi-label threshold discrimination mechanism. Differential discrimination thresholds are set according to the yin-yang syndrome type. Multi-label yin-yang syndrome identification results are generated by combining confidence information. The multi-label yin-yang syndrome identification results include syndrome type, prediction probability, and confidence.

[0052] In one specific embodiment, a threshold set for yin syndrome and a threshold set for yang syndrome are established, and corresponding probability discrimination thresholds are set for different yin and yang syndrome types; Each element in the fused prediction probability vector is compared with the corresponding discrimination threshold. When the probability value is greater than the threshold, 1 is output; otherwise, 0 is output, thus obtaining the binarized multi-label prediction result. Calculate the confidence interval for each syndrome prediction, combine the binarized prediction results, the corresponding prediction probabilities, and the confidence intervals into a structured data format, and output the multi-label yin-yang syndrome identification results. The multi-label yin-yang syndrome identification results include syndrome name, existence judgment, prediction probability value, and confidence range.

[0053] Specifically, this embodiment establishes a standardized preprocessing mechanism and a multi-label threshold discrimination mechanism to transform the probability output of the fusion model into a structured multi-label syndrome identification result, thereby realizing feature preprocessing based on Z-score standardization and improving the standardization and discrimination accuracy of the syndrome identification result.

[0054] This invention also provides a TCM Yin-Yang syndrome identification system based on the fusion of bioelectrical signal features, the system comprising: The feature extraction module is used to acquire the bioelectrical signals of the test object through remote photoplethysmography and construct a multidimensional bioelectrical signal feature vector based on the bioelectrical signals. The Yin-Yang opposition and unity constraint module is used to construct a multi-label error function for Yin-Yang opposition and unity constraint based on the Yin-Yang syndrome opposition and unity theory and the syndrome weight hierarchy theory. The Yin-Yang balance perception module is used to construct a Yin-Yang balance perception BP-MLL neural network model using the multidimensional bioelectric signal feature vector as input. The constraint-guided model training module is used to train the yin-yang balance perception BP-MLL neural network model using the yin-yang opposition and unity constraint multi-label error function to obtain the first yin-yang syndrome identification model. The machine learning model integration module is used to construct a support vector machine model and a three-layer syndrome decision tree model. The support vector machine model and the three-layer syndrome decision tree model are cascaded and integrated to obtain a second yin-yang syndrome identification model. The fusion decision module is used to fuse the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model through a confidence-weighted fusion algorithm and a dynamic rule base update mechanism to obtain a TCM theory-guided decision fusion model. The multi-label syndrome identification output module is used to input the multi-dimensional bioelectric signal feature vector into the TCM theory-guided decision fusion model for processing and output the multi-label yin-yang syndrome identification result.

[0055] Specifically, this embodiment presents a TCM Yin-Yang syndrome identification system based on fused bioelectrical signal features. By integrating traditional TCM theory with modern machine learning technology, it achieves full-process processing from bioelectrical signal acquisition, feature extraction, theoretical constraint modeling, multi-model training integration to fused decision output. This improves the objectivity of TCM Yin-Yang syndrome identification, provides a systematic, standardized, and scalable technical solution for TCM intelligent diagnosis, and promotes the transformation and upgrading of traditional TCM diagnosis towards digitalization and intelligence.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on the fusion of bioelectrical signal features, characterized in that, Includes the following steps: Bioelectric signals of the tested object are acquired by remote photoplethysmography, and a multidimensional bioelectric signal feature vector is constructed based on the bioelectric signals. Based on the theory of the unity of opposites in Yin-Yang syndromes and the theory of hierarchical weighting of syndromes, a multi-label error function constrained by the unity of opposites in Yin-Yang is constructed. Using the multidimensional bioelectric signal feature vector as input, a Yin-Yang balance perception BP-MLL neural network model is constructed; The Yin-Yang balance perception BP-MLL neural network model is trained using the Yin-Yang opposition and unity constraint multi-label error function to obtain the first Yin-Yang syndrome identification model. A support vector machine model and a three-layer syndrome decision tree model are constructed, and the support vector machine model and the three-layer syndrome decision tree model are cascaded and integrated to obtain a second yin-yang syndrome identification model. The first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model are fused together using a confidence-weighted fusion algorithm and a dynamic rule base update mechanism to obtain a TCM theory-guided decision fusion model. The multidimensional bioelectric signal feature vector is input into the TCM theory-guided decision fusion model for processing, and the multi-label yin-yang syndrome identification result is output.

2. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 1, characterized in that, The above-mentioned theory of the unity of opposites in Yin-Yang syndromes and the hierarchical theory of syndrome weights constructs a multi-label error function constrained by the unity of opposites in Yin-Yang syndromes, including: By setting a penalty function for the combination of opposing Yin and Yang syndromes, a Yin-Yang opposition constraint term is constructed to constrain the simultaneous prediction results of opposing syndromes. By setting differentiated weight coefficients based on the clinical importance level of TCM syndromes, a syndrome weight constraint term is constructed to differentiate the prediction error of primary syndrome, secondary syndrome, and concurrent syndrome. The Yin-Yang opposition constraint term and the syndrome weight constraint term are combined with the traditional multi-label loss term to obtain the Yin-Yang opposition unity constraint multi-label error function.

3. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 2, characterized in that, The penalty function for the opposing combination of Yin and Yang syndromes includes: Establish a set of Yin syndrome clusters and a set of Yang syndrome clusters, and set mutual exclusion constraint rules between Yin syndrome clusters and Yang syndrome clusters. When the output of the neural network includes the predicted values ​​of both Yin syndrome clusters and Yang syndrome clusters, input the predicted values ​​of Yin syndrome clusters and Yang syndrome clusters into a penalty function for calculation to obtain the corresponding penalty weight values. Differentiate the penalty weight values ​​from the predicted values ​​within the same type of syndrome cluster, and output the constraint term values ​​after processing by the penalty function.

4. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 1, characterized in that, The step of constructing a Yin-Yang balance perception BP-MLL neural network model using the multidimensional bioelectric signal feature vector as input includes: The Yin-Yang balance perception BP-MLL neural network model includes an input layer, a hidden layer, a Yin-Yang balance perception layer, and an output layer. A yin-yang balance perception layer is constructed, which includes a yin-yang syndrome perception mechanism. The yin-yang balance perception layer is set between the hidden layer and the output layer and is used to receive the feature representation of the hidden layer as input. A yin-yang mutual exclusion gating unit is set in the yin-yang balance perception layer. The yin-yang mutual exclusion gating unit includes a yin syndrome gating channel and a yang syndrome gating channel. The yin syndrome gating channel and the yang syndrome gating channel selectively transmit yin syndrome-related features and yang syndrome-related features, respectively. In the Yin-Yang balance perception layer, an HRV feature-guided adaptive activation function is set. The parameter configuration of the HRV feature-guided adaptive activation function is adjusted according to the type of input heart rate variability parameters. The gated features are input into the HRV feature-guided adaptive activation function for nonlinear transformation, and the processing result of the Yin-Yang balance perception layer is output.

5. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 4, characterized in that, The mutual exclusion gate unit includes: Establish a Yin syndrome feature index table and a Yang syndrome feature index table, and map the input feature vector to the Yin syndrome feature subset and the Yang syndrome feature subset respectively according to the Yin syndrome feature index table and the Yang syndrome feature index table; Set a gate weight matrix for negative syndromes and a gate weight matrix for positive syndromes, and perform matrix operations on the gate weight matrix for negative syndromes and the gate weight matrix for positive syndromes with the feature subsets for negative syndromes and the feature subsets for positive syndromes, respectively; The sigmoid gating function was used to calculate the gating values ​​for yin syndrome and yang syndrome respectively. The yin syndrome gate value and the yin syndrome feature subset are multiplied element-wise, and the yang syndrome gate value and the yang syndrome feature subset are multiplied element-wise to obtain the yin and yang syndrome feature representations after mutual exclusion gate filtering.

6. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 1, characterized in that, The process of training the Yin-Yang balance-aware BP-MLL neural network model using the Yin-Yang opposition and unity constraint multi-label error function to obtain the first Yin-Yang syndrome identification model includes: A Yin-Yang constraint gradient training mechanism is established, which decomposes the gradient information of the Yin-Yang opposition and unity constraint multi-label error function into the Yin-Yang opposition constraint gradient and the syndrome weight constraint gradient. Set a Yin-Yang balance adjustment factor, and dynamically adjust the value of the Yin-Yang balance adjustment factor according to the ratio of the prediction error of Yin syndrome to the prediction error of Yang syndrome in the current training round. The Yin-Yang balance adjustment factor is used to weight the Yin-Yang opposition constraint gradient, and the weighted Yin-Yang opposition constraint gradient is fused with the syndrome weight constraint gradient to obtain the Yin-Yang perception training gradient. The parameters of the Yin-Yang balance perception BP-MLL neural network model are updated by backpropagation using the Yin-Yang perception training gradient.

7. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 6, characterized in that, The yin-yang constraint gradient training mechanism includes: Calculate the partial derivatives of the multi-label error function constrained by the Yin-Yang opposition and unity with respect to the network parameters, and calculate the partial derivatives separately according to the Yin-Yang opposition constraint term and the syndrome weight constraint term; Establish a Yin-Yang gradient allocation matrix and allocate the gradient components of the Yin-Yang opposition constraint terms to the Yin syndrome gating channel parameters and Yang syndrome gating channel parameters of the Yin-Yang balance perception layer, respectively. The gradient components of the syndrome weight constraint term are allocated to the corresponding network layer parameters according to the weight ratio of primary syndrome, secondary syndrome, and concurrent syndrome. Set a gradient clipping threshold to restrict the numerical range of each gradient component after allocation, and output the training gradient combination after yin-yang constraint processing.

8. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 1, characterized in that, The construction of the support vector machine model and the three-layer syndrome decision tree model, and the cascading integration of the support vector machine model and the three-layer syndrome decision tree model to obtain the second yin-yang syndrome identification model, includes: A support vector machine model including constraints from traditional Chinese medicine theory is constructed. These constraints are based on the theory of Yin-Yang and Five Elements in traditional Chinese medicine to set the mutual generation and restraint relationships between syndromes. The decision boundary of the support vector machine is adjusted with theoretical guidance, and the classification results of the support vector machine are output. A three-layer syndrome decision tree model is constructed, comprising a basic judgment layer for Yin deficiency and Yang deficiency, a subdivision layer for Zang-Fu syndrome, and a syndrome severity assessment layer. The support vector machine classification results are used as inputs to the three-layer syndrome decision tree model for concatenation processing to obtain a second Yin-Yang syndrome identification model.

9. The method for identifying Yin-Yang syndromes in Traditional Chinese Medicine based on fused bioelectrical signal features as described in claim 1, characterized in that, The process of inputting the multidimensional bioelectric signal feature vector into the TCM theory-guided decision fusion model for processing and outputting multi-label Yin-Yang syndrome identification results includes: The multidimensional bioelectric signal feature vector is standardized and preprocessed. The preprocessed feature vector is then input into the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model in the TCM theory-guided decision fusion model for parallel inference calculation. The fusion prediction probability vector is obtained through a confidence-weighted fusion algorithm. The fused prediction probability vector is binarized using a multi-label threshold discrimination mechanism. Differential discrimination thresholds are set according to the yin-yang syndrome type. Multi-label yin-yang syndrome identification results are generated by combining confidence information. The multi-label yin-yang syndrome identification results include syndrome type, prediction probability, and confidence.

10. A TCM Yin-Yang syndrome identification system based on fused bioelectrical signal features, used to execute the TCM Yin-Yang syndrome identification method based on fused bioelectrical signal features as described in any one of claims 1-9, characterized in that, The system includes: The feature extraction module is used to acquire the bioelectrical signals of the test object through remote photoplethysmography and construct a multidimensional bioelectrical signal feature vector based on the bioelectrical signals. The Yin-Yang opposition and unity constraint module is used to construct a multi-label error function for Yin-Yang opposition and unity constraint based on the Yin-Yang syndrome opposition and unity theory and the syndrome weight hierarchy theory. The Yin-Yang balance perception module is used to construct a Yin-Yang balance perception BP-MLL neural network model using the multidimensional bioelectric signal feature vector as input. The constraint-guided model training module is used to train the yin-yang balance perception BP-MLL neural network model using the yin-yang opposition and unity constraint multi-label error function to obtain the first yin-yang syndrome identification model. The machine learning model integration module is used to construct a support vector machine model and a three-layer syndrome decision tree model. The support vector machine model and the three-layer syndrome decision tree model are cascaded and integrated to obtain a second yin-yang syndrome identification model. The fusion decision module is used to fuse the first Yin-Yang syndrome identification model and the second Yin-Yang syndrome identification model through a confidence-weighted fusion algorithm and a dynamic rule base update mechanism to obtain a TCM theory-guided decision fusion model. The multi-label syndrome identification output module is used to input the multi-dimensional bioelectric signal feature vector into the TCM theory-guided decision fusion model for processing and output the multi-label yin-yang syndrome identification result.

Images