Fuzzy theory fused tongue diagnosis characteristic quantitative analysis method and system

Abstract

The invention discloses a fuzzy theory fused tongue diagnosis characteristic quantitative analysis method and system, and relates to the technical field of traditional Chinese medicine tongue diagnos.The method comprises the steps that a micro piezoelectric sensor array and a hyperspectral imaging module are integrated, tongue surface Young modulus distribution data and hyperspectral reflectivity data are synchronously collected, and a multi-mode tongue picture fusion data set is generated; further performing crack region segmentation and super-resolution reconstruction through a neural network model, outputting a crack network diagram and a tongue papilla form entropy, generating topological fingerprints in combination with continuous homology analysis, identifying a mechanical gradient mutation region and extracting a metabolic index, constructing a mechanical-metabolic associated feature set, and obtaining a mechanical-metabolic associated feature set; topological fingerprints and mechanical-metabolic characteristics are fused in a three-dimensional space-time coordinate system, a dynamic intuitionistic fuzzy decision field is constructed, finally, tongue picture pathology evolution risk indexes and a visual early warning report are output through a pathology prediction model, and the system provides quantifiable decision support for'prevention before illness' in traditional Chinese medicine.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traditional Chinese medicine tongue diagnosis, and particularly relates to a tongue diagnosis feature quantitative analysis method and system fusing fuzzy theory. BACKGROUND

[0002] In recent years, the objective technology of traditional Chinese medicine tongue diagnosis gradually integrates computer vision and artificial intelligence methods. The mainstream scheme segments the tongue fur / tongue body area through RGB image, and establishes a diagnosis rule base combined with fuzzy logic. Hyperspectral imaging technology is introduced to improve the analysis accuracy of tongue color, and some studies attempt to extract crack morphological features. With the development of flexible sensing technology, piezoelectric sensors are used for oral physiological parameter monitoring, but have not been deeply integrated with tongue image analysis systems. The existing technology focuses on single modal data processing, and lacks a collaborative analysis framework for multi-source heterogeneous features.

[0003] Traditional methods have many limitations: first, the dependence on optical images leads to the lack of mechanical properties, and the gradient change of the elastic modulus of the tongue body cannot be quantified; second, crack analysis only stays at the morphological segmentation level, lacks topological invariant representation, and makes the quantification of complex crack networks rough; third, the static fuzzy rule base cannot capture the spatiotemporal dynamic evolution features, and cannot construct a cross-scale decision field of metabolism-mechanics-topology. The above defects lead to the insufficient sensitivity of the existing system in early pathological evolution warning, and it is difficult to meet the clinical needs of traditional Chinese medicine "prevention before disease". SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a tongue diagnosis feature quantitative analysis method fusing fuzzy theory to solve the problems of the existing tongue diagnosis feature quantitative analysis method, such as the lack of multi-modal data fusion leading to the lack of biomechanical features, the crack quantification relying on morphological segmentation and lacking topological invariant analysis, and how to construct a spatiotemporal dynamic intuitionistic fuzzy decision field to realize early pathological warning.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a tongue diagnosis feature quantitative analysis method fusing fuzzy theory, characterized in that it comprises the following steps:

[0008] A micro-piezoelectric sensor array and a hyperspectral imaging module integrated in a tongue diagnosis device are used to synchronously collect tongue surface Young's modulus distribution data and hyperspectral reflectance data, and generate a multi-modal tongue image fusion data set;

[0009] The hyperspectral reflectance data is input into a neural network model for crack region segmentation and super-resolution reconstruction, and the crack network graph with geometric topological structure and tongue papilla morphological entropy are output.

[0010] performing continuous homology analysis on the crack network graph, calculating crack topological invariants and generating a topological fingerprint of quantitative crack complexity;

[0011] identifying a mechanical gradient mutation zone based on the Young's modulus distribution data, extracting a metabolic index in combination with hyperspectral reflectance, and constructing a mechanical-metabolic correlation feature set;

[0012] fusing the topological fingerprint and the mechanical-metabolic correlation feature set in a three-dimensional spatiotemporal coordinate system to construct a dynamic intuitionistic fuzzy decision field;

[0013] inputting the intuitionistic fuzzy decision field into a pathological prediction model to output a tongue image pathological evolution risk index and a visual warning report.

[0014] As a preferred scheme of the tongue diagnosis feature quantitative analysis method based on fuzzy theory, the micro piezoelectric sensor array and the hyperspectral imaging module integrated in the tongue diagnosis device are used to synchronously collect the tongue surface Young's modulus distribution data and the hyperspectral reflectance data, and generate a multimodal tongue image fusion data set, and the specific steps are as follows:

[0015] The micro piezoelectric sensor array is arranged in a ring topology to apply gradient pressure: 0-5 kPa in the center area, 5-10 kPa in the transition area, and 10-15 kPa in the edge area.

[0016] During the pressure loading process, the hyperspectral imaging module captures reflectance data at 5 wavelengths (540 nm, 560 nm, 920 nm, 970 nm, and 620 nm).

[0017] The dynamic feedback adjustment module is used to real-time calibrate the pressure-deformation curve to eliminate tongue body movement artifacts.

[0018] The mechanical response data and the spectral data are spatiotemporally aligned to generate a three-dimensional coordinate-bound fusion data set.

[0019] As a preferred scheme of the tongue diagnosis feature quantitative analysis method based on fuzzy theory, the hyperspectral reflectance data is input into a neural network model for crack region segmentation and super-resolution reconstruction, and the crack network graph and the papilla shape entropy with geometric topological structure are output, and the specific steps are as follows:

[0020] In the HSV color space, candidate crack regions with saturation>0.7 and brightness<0.4 are extracted.

[0021] A non-local attention U-Net++ model is used to perform sub-pixel level segmentation on the candidate regions, and a binary crack mask is output.

[0022] The mask is converted into a topological graph structure based on a skeletonization algorithm, where the intersection points are vertices and the crack paths are edges.

[0023] A curvature histogram of the tongue papilla profile is calculated to generate an entropy value representing the morphological complexity.

[0024] As a preferred scheme of the tongue diagnosis feature quantification analysis method of fusing fuzzy theory, wherein: the crack network graph is subjected to persistent homology analysis, crack topological invariants are calculated, and a topological fingerprint quantifying crack complexity is generated, the specific steps being,

[0025] The vertex set of the crack topological graph is mapped into two-dimensional point cloud data;

[0026] A scale-increasing topological structure is constructed through Vietoris-Rips complex filtering;

[0027] A barcode set with a survival interval exceeding 0.15 mm is extracted from the one-dimensional homology group;

[0028] Betti numbers and persistent entropy of the barcode are calculated as the topological fingerprint.

[0029] As a preferred scheme of the tongue diagnosis feature quantification analysis method of fusing fuzzy theory, wherein: based on the Young's modulus distribution data, a mechanical gradient mutation zone is identified, a metabolic index is extracted in combination with hyperspectral reflectance, and a mechanical-metabolic correlation feature set is constructed, the specific steps being,

[0030] The Young's modulus distribution is subjected to anisotropic diffusion filtering, and a gradient amplitude is calculated;

[0031] Regions with a gradient amplitude greater than 2.5 kPa / mm are marked as mechanical mutation zones;

[0032] Oxygenation and hydration indices are extracted at 540 nm and 970 nm wavebands;

[0033] A weighted mapping matrix of the mechanical gradient in the mutation zone and the metabolic index is established.

[0034] As a preferred scheme of the tongue diagnosis feature quantification analysis method of fusing fuzzy theory, wherein: the topological fingerprint and the mechanical-metabolic correlation feature set are fused in a three-dimensional spatiotemporal coordinate system to construct a dynamic intuitionistic fuzzy decision field, the specific steps being,

[0035] In a three-dimensional coordinate system, spatial dimensions (x, y) and a time stamp t are defined;

[0036] Betti numbers of the topological fingerprint are converted into crack membership functions;

[0037] Mechanical gradient amplitudes are converted into tissue elasticity non-membership functions;

[0038] Metabolic conflict indices are fused to generate a spatiotemporal dynamic hesitancy field.

[0039] As a preferred scheme of the tongue diagnosis feature quantitative analysis method fusing fuzzy theory, wherein: the intuitive fuzzy decision field is input into a pathology prediction model, and a tongue image pathology evolution risk index and a visual warning report are output, and the specific steps are,

[0040] Discretize the fuzzy decision field into a space-time grid node;

[0041] Based on Lyapunov index calibration, the chaotic parameter prediction field is evolved;

[0042] Render the pathology risk thermodynamic map in the hyperbolic Poincare disc;

[0043] Integrate the risk gradient to generate an early lesion warning index.

[0044] In a second aspect, the present application provides a tongue diagnosis feature quantitative analysis system fusing fuzzy theory, comprising: a multi-modal acquisition module, which synchronously acquires tongue surface Young's modulus distribution data and hyperspectral reflectance data through a micro-piezoelectric sensor array and a hyperspectral imaging module integrated in a tongue diagnosis device, and generates a multi-modal tongue image fusion data set;

[0045] A crack reconstruction module, which inputs the hyperspectral reflectance data into a neural network model for crack area segmentation and super-resolution reconstruction, and outputs a crack network graph with geometric topological structure and a tongue papilla morphological entropy;

[0046] A topological quantification module, which performs persistent homology analysis on the crack network graph, calculates crack topological invariants, and generates a topological fingerprint of quantitative crack complexity;

[0047] A cross-scale correlation module, which identifies a mechanical gradient mutation zone based on the Young's modulus distribution data, extracts a metabolic index in combination with the hyperspectral reflectance, and constructs a mechanical-metabolic correlation feature set;

[0048] A dynamic decision module, which fuses the topological fingerprint and the mechanical-metabolic correlation feature set in a three-dimensional space-time coordinate system, and constructs a dynamic intuitive fuzzy decision field;

[0049] A risk warning module, which inputs the intuitive fuzzy decision field into a pathology prediction model, and outputs a tongue image pathology evolution risk index and a visual warning report.

[0050] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, any step of the tongue diagnosis feature quantitative analysis method fusing fuzzy theory according to the first aspect of the present application is implemented.

[0051] In a fourth aspect, the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements any step of the tongue diagnosis feature quantification analysis method based on fuzzy theory according to the first aspect of the present application.

[0052] The present application has the following beneficial effects: through multi-modal synchronous acquisition of micro-piezoelectric sensors and hyperspectral imaging, the tongue surface biomechanical properties and optical features are cooperatively quantified, breaking through the limitation of traditional tongue diagnosis relying on single image data; the topological fingerprint is generated based on the persistent homology analysis of the crack network, converting the subjective morphological description into objective topological invariant, and significantly improving the quantification accuracy of complex cracks; the cross-scale mechanical-metabolic correlation feature set is constructed, the mapping relationship between the tongue body elastic modulus gradient and metabolic indicators is established, and the deep pathological mechanism of the "shape-quality" correlation is revealed; the dynamic intuitionistic fuzzy decision field is formed in the three-dimensional space-time coordinate system, the topological fingerprint and mechanical-metabolic features are fused, and the uncertainty of pathological evolution is captured by combining the hesitation operator, realizing the leap from static rules to dynamic early warning; finally, through the hyperbolic space visualization driven by chaos, the early lesion risk thermal map is output, providing quantifiable decision support for the traditional Chinese medicine "prevention before disease". BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. 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.

[0054] Figure 1 The flowchart of the tongue diagnosis feature quantification analysis method based on fuzzy theory in Example 1. DETAILED DESCRIPTION

[0055] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0056] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0057] Second, the "one embodiment" or "an embodiment" referred to herein can include a particular feature, structure, or characteristic. The various embodiments appearing at different places in this specification are not necessarily all cumulative or alternative implementations of the application. In other words, the "in one embodiment" appearing at different places in this specification are not necessarily all referring to the same embodiment or a single alternative implementation of the application.

[0058] Embodiment 1, Reference Figure 1 For the first embodiment of the application, the embodiment provides a tongue diagnosis feature quantification analysis method integrating fuzzy theory, characterized by comprising the following steps:

[0059] Through the micro piezoelectric sensor array and the hyperspectral imaging module integrated in the tongue diagnosis device, the Young's modulus distribution data and the hyperspectral reflectance data of the tongue surface are synchronously collected to generate a multi-modal tongue image fusion data set;

[0060] The hyperspectral reflectance data is input into a neural network model for crack region segmentation and super-resolution reconstruction, and a crack network graph with geometric topological structure and a tongue papilla morphological entropy are output;

[0061] The crack network graph is subjected to persistent homology analysis, and a topological fingerprint of quantified crack complexity is generated by calculating crack topological invariants;

[0062] Based on the Young's modulus distribution data, a mechanical gradient mutation area is identified, and a metabolic index is extracted in combination with the hyperspectral reflectance to construct a mechanical-metabolic correlation feature set;

[0063] The topological fingerprint and the mechanical-metabolic correlation feature set are fused in a three-dimensional spatiotemporal coordinate system to construct a dynamic intuitionistic fuzzy decision field;

[0064] The intuitionistic fuzzy decision field is input into a pathological prediction model to output a tongue image pathological evolution risk index and a visualized early warning report.

[0065] It should be noted that the multi-modal synchronous collection

[0066] A gradient pressure is applied by the annularly arranged micro piezoelectric sensor array (0-5 kPa in the center area, 5-10 kPa in the transition area, and 10-15 kPa in the edge area), and the hyperspectral imaging module is triggered to synchronously capture reflectance data at 540 nm, 560 nm, 920 nm, 970 nm, and 620 nm during the pressure loading process. A dynamic feedback adjustment module is used to real-time calibrate the pressure-deformation curve to eliminate tongue displacement interference, and finally the mechanical response data and the spectral data are aligned according to the three-dimensional spatial coordinates to generate a fusion data set.

[0067] By gradient pressure loading and multi-band synchronous capture mechanism, the spatiotemporal alignment of tongue surface biomechanical properties (Young's modulus) and metabolic optical features (reflectance) is realized for the first time, solving the problem of missing mechanical features caused by the reliance on single optical data in traditional tongue diagnosis, and providing data basis for cross-scale correlation analysis.

[0068] Crack topology reconstruction

[0069] The candidate crack region with saturation >0.7 and brightness <0.4 in HSV color space is screened, and the sub-pixel level segmentation is generated by inputting the non-local attention U-Net++ model to generate the binary mask. Based on the skeletonization algorithm, the mask is converted into a topological graph structure (vertex = crack intersection, edge = crack path), and the curvature histogram entropy value of the papilla contour is calculated The non-local attention mechanism and skeletonization algorithm are used to convert the crack morphology into a geometric topological graph structure and quantify the papilla curvature entropy, breaking through the limitations of traditional morphological segmentation for broken cracks and providing structured input for topological invariant calculation.

[0070] The vertex set of the crack topological graph is mapped to a two-dimensional point cloud The Vietoris-Rips complex filter is used to construct a scale-increasing topological structure, and the barcode of the survival interval of the one-dimensional homology group H1 exceeding 0.15 mm is extracted Betti number β1 and persistent entropy are calculated as topological fingerprints.

[0071] Based on the persistent homology theory, the crack network is abstracted into topological invariants β1 and S p , establishing a quantifiable crack complexity diagnostic standard, overcoming the subjectivity of manual interpretation and the lack of representation ability of complex topological structure in traditional pixel statistical methods.

[0072] Cross-scale correlation modeling

[0073] The gradient amplitude is calculated after anisotropic diffusion filtering of the Young's modulus distribution E(x, y) The area with gradient >2.5 kPa / mm is marked as the mechanical mutation area, and the oxygenation index is extracted at 540 nm and 970 nm wavebands respectively and hydration index A weighted mapping matrix M is constructed for the mechanical gradient and metabolic index of the mutation area ij .

[0074] By identifying the mechanical gradient mutation area and correlating the spectral metabolic index, the cross-scale mapping relationship between tongue elasticity and biochemical metabolism is established, revealing the quantitative evidence of the "shape-quality correlation" theory in traditional Chinese medicine, and making up for the one-sidedness of single modal feature analysis.

[0075] Dynamic decision field construction

[0076] In the three-dimensional coordinate system (x, y, t), the spatial-time dimension is defined, the β1 value of the topological fingerprint is converted into the crack membership function, the mechanical gradient amplitude is converted into the tissue elasticity non-membership function, and the metabolic conflict index is fused to generate an intuitionistic fuzzy field μIFS = <μ, v, π>.

[0077] In the space-time coordinate system, the topological, mechanical and metabolic characteristics are fused to construct a dynamic intuitionistic fuzzy field, the hesitancy degree operator π is used to quantify the diagnosis uncertainty, the leap from the static rule base to the dynamic evolving decision is realized, and the problem that the traditional fuzzy system is difficult to capture the pathological process is solved.

[0078] Pathological early warning output

[0079] The intuitionistic fuzzy field Φ is discretized into a space-time grid node, a chaotic dynamics equation is established based on the Lyapunov index calibration parameter , a risk thermodynamic map is rendered in the hyperbolic Poincare disc, and an early warning index is generated by integrating the risk gradient .

[0080] The combination of chaotic dynamics and hyperbolic geometry realizes the spatio-temporal evolution prediction and visualization of pathological risk, outputs a quantitative early warning index of early lesions, provides an operable decision-making tool for the traditional Chinese medicine "prevention before disease", and breaks through the limitation of the prior art that only static diagnostic conclusions can be output.

[0081] Specifically, the micro piezoelectric sensor array and the hyperspectral imaging module integrated in the tongue diagnosis device are used to synchronously collect the Young's modulus distribution data and the hyperspectral reflectance data of the tongue surface, and generate a multi-modal tongue image fusion data set, and the specific steps are as follows,

[0082] The micro piezoelectric sensor array is arranged in a ring topology to apply gradient pressure: 0-5kPa in the center area, 5-10kPa in the transition area, and 10-15kPa in the edge area.

[0083] During the pressure loading process, the hyperspectral imaging module captures reflectance data at 5 wavebands (540nm, 560nm, 920nm, 970nm, 620nm);

[0084] The dynamic feedback adjustment module is used to real-time calibrate the pressure-deformation curve and eliminate tongue movement artifacts.

[0085] The mechanical response data and the spectral data are spatio-temporally aligned to generate a three-dimensional coordinate bound fusion data set.

[0086] It should be noted that the micro piezoelectric sensor array is arranged in a ring topology to apply gradient pressure: 0-5kPa in the center area, 5-10kPa in the transition area, and 10-15kPa in the edge area.

[0087] Firstly, the micro piezoelectric sensor array is precisely arranged and designed to be laid out in a ring topology. This layout ensures that the pressure application has a clear regional division and gradient change. In actual operation, the output pressure of each sensor is precisely adjusted through control signals, so that the pressure range of the central area is controlled at 0-5kPa, the pressure range of the transition area is 5-10kPa, and the pressure range of the edge area is 10-15kPa. This gradient pressure application method can simulate the different mechanical stimuli received by the tongue body in different areas, so as to more accurately obtain the mechanical response characteristics of the tongue body.

[0088] Through this ring topology arrangement and gradient pressure application method, the mechanical characteristics of different areas of the tongue body are precisely detected. The lower pressure in the central area can detect the softness of the core area of the tongue body, the pressure change in the transition area can capture the mechanical gradient change of the tongue body from the center to the edge, and the higher pressure in the edge area can detect the mechanical characteristics of the edge of the tongue body. This regional pressure application method can more comprehensively reflect the mechanical state of the tongue body, providing a richer data basis for subsequent analysis.

[0089] Trigger the hyperspectral imaging module to capture reflectance data at 5 bands (540nm, 560nm, 920nm, 970nm, 620nm) during the pressure loading process

[0090] When the micro piezoelectric sensor array starts to apply gradient pressure, the hyperspectral imaging module is started simultaneously. This module can capture reflectance data on the surface of the tongue body within a specific band range. Specifically, the selected 5 bands are 540nm, 560nm, 920nm, 970nm and 620nm. These bands cover the range from visible light to near-infrared, which can reflect the optical characteristics of the tongue body under different spectra. The imaging module uses high-precision optical sensors and spectral filters to quickly scan the surface of the tongue body and obtain reflectance data at each band. These data can reflect the pigment distribution, blood vessel condition and optical characteristics of the tissue on the surface of the tongue body.

[0091] By synchronously capturing hyperspectral reflectance data during the pressure loading process, real-time correlation of mechanical stimulation and optical response is achieved. This multi-modal data acquisition method can more comprehensively reflect the physiological changes of the tongue body under mechanical stimulation. For example, reflectance data at different bands can be used to analyze blood flow changes, pigment deposition and metabolic state of the tissue on the surface of the tongue body. The combination of multi-modal data provides more information for subsequent tongue analysis, which helps to improve the accuracy and reliability of diagnosis.

[0092] Real-time calibration of pressure-deformation curve through dynamic feedback adjustment module to eliminate tongue movement artifacts

[0093] During the pressure loading and data collection process, the tongue body may move slightly due to physiological reactions or external disturbances, which may cause artifacts in the collected data and affect the accuracy of analysis. Therefore, a dynamic feedback adjustment module is introduced. This module monitors the position changes of the tongue body and the pressure-deformation curve in real time, and uses feedback algorithms to dynamically adjust the pressure application process. Specifically, when the tongue body is detected to move, the module quickly adjusts the output pressure of the micro piezoelectric sensor to compensate for the deformation changes caused by the tongue body movement. At the same time, through the optimization algorithm, the pressure-deformation curve is calibrated in real time to ensure that the collected data can truly reflect the mechanical response of the tongue body.

[0094] Through the real-time calibration of the dynamic feedback adjustment module, the artifacts caused by the tongue body movement are effectively eliminated, improving the accuracy and reliability of data collection. This real-time calibration mechanism ensures the accuracy of the pressure-deformation curve, enabling subsequent analysis to be based on high-quality data. This is of great significance to improving the precision and stability of tongue diagnosis, especially in clinical applications, which can reduce the possibility of misdiagnosis.

[0095] Time and space alignment of mechanical response data and spectral data to generate a three-dimensional coordinate-bound fusion data set

[0096] After the collection of mechanical response data and spectral data, the two types of data need to be time and space aligned. Specifically, through accurate time synchronization and spatial positioning algorithms, the mechanical response data at each time point and spatial position is matched with the corresponding spectral data. This process involves accurate processing of the timestamps and spatial coordinates of the data to ensure their consistency in time and space. Finally, a three-dimensional coordinate-bound fusion data set is generated, which contains mechanical response and spectral information at each data point, forming a multi-modal, multi-dimensional data set.

[0097] The three-dimensional coordinate-bound fusion data set generated through time and space alignment realizes the deep fusion of mechanical and spectral data. This fusion data set can more comprehensively reflect the physiological changes of the tongue under mechanical stimulation, providing more abundant information for subsequent analysis. For example, the changes in mechanical response and spectral characteristics of different regions of the tongue can be observed intuitively in three-dimensional space, which helps to discover potential pathological features. The fusion of multi-modal data provides a more powerful tool for tongue diagnosis, which can improve the accuracy and efficiency of diagnosis.

[0098] Specifically, the hyperspectral reflectance data is input into the neural network model for crack region segmentation and super-resolution reconstruction, and the crack network graph with geometric topology structure and the tongue papilla morphological entropy are output, and the specific steps are,

[0099] candidate crack regions with saturation > 0.7 and lightness < 0.4 are extracted in HSV color space;

[0100] A non-local attention U-Net++ model is used to perform sub-pixel level segmentation on the candidate regions, and a binary crack mask is output;

[0101] The mask is converted into a topological graph structure based on a skeletonization algorithm, where the intersection points are vertices and the crack paths are edges.

[0102] The curvature histogram of the papilla contour is calculated to generate an entropy value representing the morphological complexity.

[0103] It should be noted that candidate crack regions with saturation > 0.7 and lightness < 0.4 are extracted in HSV color space

[0104] First, the tongue image is converted from RGB color space to HSV color space, because HSV color space can more intuitively represent the saturation and lightness of colors. In HSV color space, saturation represents the purity of color, and lightness represents the brightness of color. By analyzing the tongue image pixel by pixel, the pixel points that meet the condition of saturation greater than 0.7 and lightness less than 0.4 are selected. The region composed of these pixel points is the candidate crack region.

[0105] By extracting regions with specific saturation and lightness ranges in HSV color space, the crack region can be effectively separated from the complex tongue background. This method takes advantage of the color characteristics of the crack region, which is usually dark and has high color purity, thereby improving the accuracy and efficiency of crack detection. It provides a clearer and more accurate target region for subsequent crack segmentation and analysis, reducing the complexity and computational load of subsequent processing.

[0106] A non-local attention U-Net++ model is used to perform sub-pixel level segmentation on the candidate regions, and a binary crack mask is output

[0107] The non-local attention U-Net++ model is used to process the candidate crack regions extracted in the first step. This model is a deep learning-based image segmentation network that combines the advantages of non-local attention mechanism and U-Net++ architecture. During model training, a large number of labeled tongue images are used as training data to enable the model to learn the features and patterns of crack regions. In actual application, the candidate crack regions are input into the model, which performs sub-pixel level segmentation on the crack regions through multi-scale feature extraction and non-local attention modules, and finally outputs a binary crack mask.

[0108] The non-local attention U-Net++ model can perform high-precision sub-pixel level segmentation on crack regions, effectively improving the resolution and accuracy of crack detection. This model combines the non-local attention mechanism, which can better capture the long-distance dependencies and detailed features of crack regions, thus more accurately segmenting the boundaries of cracks. The output binary crack mask provides a clear and accurate basis for subsequent crack analysis and feature extraction, helping to improve the reliability of tongue pathology analysis.

[0109] The mask is converted into a topological graph structure based on the skeletonization algorithm, where the intersection points are vertices and the crack paths are edges

[0110] Based on the binary crack mask output in the second step, a skeletonization algorithm is applied for processing. Skeletonization is an image processing method that converts the target region in a binary image into a single-pixel width skeleton. Through this algorithm, the crack region in the crack mask is converted into a skeletonized line, which is then further converted into a topological graph structure. In the topological graph, the intersection points of the crack paths are defined as vertices, and the crack paths themselves are defined as edges.

[0111] Converting the crack mask into a topological graph structure allows for the description and analysis of crack networks from a topological perspective. This topological graph structure can clearly represent the connection relationships and path information between cracks, facilitating subsequent quantitative analysis of the complexity and connectivity of crack networks. By defining the intersection points of crack paths as vertices and the crack paths as edges, the topological structure of the crack network can be more intuitively represented, providing more information and powerful tools for tongue pathology analysis.

[0112] Calculate the curvature histogram of the tongue papilla contour to generate an entropy value representing the morphological complexity

[0113] First, the tongue papilla contour is extracted to obtain the boundary curve of the tongue papilla. Then, the curvature value of each point along the boundary curve is calculated. Curvature is a geometric quantity that describes the degree of bending of a curve. By calculating the curvature of each point on the tongue papilla contour, the curvature distribution can be obtained. Next, statistical analysis is performed on the curvature values to generate a curvature histogram. The curvature histogram can intuitively reflect the distribution of the curvature of the tongue papilla contour, including the range and frequency of the curvature. Finally, the entropy value representing the morphological complexity is calculated based on the curvature histogram.

[0114] By calculating the curvature histogram of the tongue papilla contour and generating the entropy value, the complexity of the tongue papilla morphology can be quantitatively described. The change in the morphology of the tongue papilla is closely related to the pathological state of the tongue, and this quantitative description provides an important indicator for tongue pathology analysis. The entropy value can reflect the irregularity of the tongue papilla contour, and a higher entropy value usually means more complex morphological features.

[0115] Specifically, the crack network graph is subjected to persistent homology analysis, crack topological invariants are calculated, and a topological fingerprint quantifying crack complexity is generated, the specific steps being,

[0116] Mapping the vertex set of the crack topological graph to two-dimensional point cloud data;

[0117] Constructing scale-increasing topological structures through Vietoris-Rips complex filtering;

[0118] Extracting a barcode set with a survival interval exceeding 0.15 mm from the one-dimensional homology group;

[0119] Calculating the Betti number and persistent entropy of the barcode as a topological fingerprint.

[0120] It should be noted that the vertex set of the crack topological graph is mapped to two-dimensional point cloud data

[0121] First, all vertex information in the crack topological graph is extracted, and these vertices represent the intersection points of the crack paths and are the key feature points of the crack network. Then, according to the position coordinates of each vertex in the topological graph, it is mapped to a two-dimensional plane to form point cloud data. This process involves coordinate conversion and data format adjustment to ensure that the two-dimensional coordinates of each vertex accurately reflect its relative position in the crack network.

[0122] By mapping the vertex set of the crack topological graph to two-dimensional point cloud data, the visualization and quantitative representation of the feature points of the crack network are achieved. This representation facilitates subsequent mathematical processing and analysis, providing a basis for the complexity research of the crack network. At the same time, two-dimensional point cloud data can intuitively show the distribution characteristics of the crack network, which helps to discover potential pathological patterns.

[0123] Constructing scale-increasing topological structures through Vietoris-Rips complex filtering

[0124] Using Vietoris-Rips complex filtering technology, the two-dimensional point cloud data generated in the first step is processed. This technology gradually constructs the topological structure of the point cloud data by setting different distance thresholds. Specifically, starting from a small distance threshold, the threshold is gradually increased, and each increase will connect points with a distance less than the current threshold to form a more complex topological structure. This process can generate a series of topological structures of different scales, reflecting the connection relationships of the crack network at different scales.

[0125] By constructing scale-increasing topological structures through Vietoris-Rips complex filtering, the topological characteristics of the crack network at different scales can be comprehensively captured. This multi-scale analysis method helps to reveal the complexity and hierarchy of the crack network, providing more abundant information for subsequent topological analysis. At the same time, this structure can better reflect the global and local characteristics of the crack network, helping to discover potential pathological changes.

[0126] Extracting a set of barcodes with a survival interval exceeding 0.15mm from the one-dimensional homology group

[0127] Specific operation: In the one-dimensional homology group, barcodes are tools for indicating the existence and disappearance of ring structures in the topological structure. By analyzing the multi-scale topological structure generated in the second step, a set of barcodes with a survival interval (i.e., the length of the barcode) exceeding 0.15mm is extracted. Specifically, the survival interval of each barcode is calculated, and barcodes with a length greater than 0.15mm are selected as the analysis object. This process involves in-depth analysis and feature extraction of the topological structure.

[0128] Extracting a set of barcodes with a survival interval exceeding 0.15mm can effectively filter out crack ring structures with significant topological characteristics. These barcodes represent relatively stable ring features in the crack network, reflecting the complexity and stability of the crack network. By focusing on these significant topological features, the pathological characteristics of the crack network can be more accurately analyzed, providing strong support for disease diagnosis and research.

[0129] Calculating the Betti number and persistent entropy of the barcodes as topological fingerprints

[0130] For the set of barcodes extracted in the third step, the Betti number and persistent entropy of each barcode are calculated. Betti number is a topological index used to describe the number of connected components, loops and voids in the topological structure, while persistent entropy is an index that measures the complexity and stability of the topological structure. By calculating these indicators, the characteristics of each barcode are quantified to form topological fingerprints. This process involves mathematical analysis and feature extraction of the topological structure.

[0131] Calculating the Betti number and persistent entropy of the barcodes as topological fingerprints can quantitatively describe and uniquely identify the topological characteristics of the crack network. This topological fingerprint can reflect the complexity and stability of the crack network, providing standardized indicators for the analysis and comparison of crack networks. Through this quantitative method, the characteristics of crack networks in different tongue images can be more intuitively compared, helping to discover potential pathological change rules and providing a new perspective and method for tongue pathology research.

[0132] Specifically, based on the Young's modulus distribution data, the mechanical gradient mutation area is identified, and the metabolism index is extracted combined with the hyperspectral reflectance to construct a mechanical-metabolic correlation feature set, the specific steps are,

[0133] Anisotropic diffusion filtering is performed on the Young's modulus distribution to calculate the gradient amplitude;

[0134] Regions with gradient amplitude > 2.5 kPa / mm are marked as mechanical mutation areas;

[0135] Oxygenation index and hydration index are extracted at 540 nm and 970 nm wavebands;

[0136] A weighted mapping matrix of mechanical gradient and metabolic index in the mutation area is established.

[0137] It should be noted that anisotropic diffusion filtering is performed on the Young's modulus distribution to calculate the gradient amplitude

[0138] First, the Young's modulus distribution data of the tongue image are obtained, which reflect the elastic properties of different regions of the tongue surface. Then, anisotropic diffusion filtering algorithm is applied to these data. This algorithm is an image processing technique that can smooth the image while preserving important edge information. During the processing, the filtering parameters are dynamically adjusted according to the local characteristics of the Young's modulus distribution, so that the filtering process has different smoothing degrees in different directions. In this way, noise can be effectively removed while preserving the key features of the Young's modulus distribution. Finally, the gradient amplitude of the filtered Young's modulus distribution data is calculated, which reflects the rate of change of the Young's modulus in space and can identify areas with dramatic changes in mechanical properties.

[0139] By anisotropic diffusion filtering of the Young's modulus distribution data, noise can be effectively removed while important mechanical features are preserved. This makes the subsequent gradient amplitude calculation more accurate, allowing more clear identification of areas with dramatic changes in mechanical properties. These areas may be closely related to the pathological state of the tongue image, so this step provides an accurate data basis for subsequent identification of mechanical mutation areas, helping to improve the reliability of tongue image pathological analysis.

[0140] Regions with gradient amplitude > 2.5 kPa / mm are marked as mechanical mutation areas

[0141] Based on the gradient amplitude data calculated in the first step, a threshold of 2.5 kPa / mm is set. Then, the entire gradient amplitude data is scanned, and all regions with gradient amplitude exceeding the threshold are marked as mechanical mutation areas. These areas represent a significant change in the Young's modulus in space, which may indicate a mutation in the mechanical properties of the tongue image tissue. In this way, areas with potential pathological significance in the tongue image can be accurately located.

[0142] By setting clear threshold values to mark the mechanical mutation zones, the areas where the mechanical properties of the tongue image change significantly can be quickly and accurately identified. These areas are often crucial for pathological analysis of the tongue image, as they may be closely related to the pathological sites of the tongue image. This step enables precise identification of the mechanical mutation zones, providing clear target areas for subsequent metabolic index analysis and pathological diagnosis, which helps to improve the accuracy and efficiency of diagnosis.

[0143] Extracting oxygenation index and hydration index at 540nm and 970nm wavebands

[0144] Hyperspectral imaging technology is used to obtain the reflectance data of the tongue image at different wavebands. Among these data, special attention is paid to the 540nm and 970nm wavebands. The reflectance data at the 540nm waveband is mainly used to extract the oxygenation index, which reflects the degree of oxygenation of the blood in the tongue tissue; while the reflectance data at the 970nm waveband is used to extract the hydration index, which reflects the water content of the tongue tissue. Through specific algorithms, the oxygenation index and hydration index are calculated from the data at these two wavebands, obtaining quantitative information about the tongue tissue in these two important physiological parameters.

[0145] By extracting the oxygenation index and hydration index at specific wavebands, two important physiological parameters of the tongue tissue can be obtained. The oxygenation index reflects the oxygenation of the blood in the tongue tissue, while the hydration index reflects the water content of the tissue. These two parameters are of great significance for pathological analysis of the tongue image, as their changes are often closely related to the pathological state of the tongue image. This step provides important physiological parameters for subsequent mechanical-metabolic correlation analysis, helping to more comprehensively understand the pathological characteristics of the tongue image.

[0146] Establishing a weighted mapping matrix of mechanical gradient and metabolic index in mutation zones

[0147] On the basis of the previous steps, the gradient amplitude data of the marked mechanical mutation zones are integrated with the extracted oxygenation index and hydration index data. First, determine the corresponding oxygenation index and hydration index values for each mechanical mutation zone. Then, according to the correlation between these data, a weighted mapping matrix is established. This matrix quantitatively represents the relationship between mechanical gradient and metabolic index by assigning different weights. The determination of weights can be based on experimental data or existing physiological models to ensure that the mapping matrix accurately reflects the intrinsic relationship between the two.

[0148] By establishing a weighted mapping matrix, a quantitative correlation between mechanical gradients and metabolic indices is achieved. This correlation can more comprehensively reflect the mechanical and physiological characteristics of tongue tissue, providing a comprehensive index for tongue pathology analysis. This step helps to reveal the potential link between mechanical mutation zones and metabolic changes, thereby providing more accurate and comprehensive information support for tongue pathology diagnosis, which helps to improve the accuracy and reliability of diagnosis.

[0149] Specifically, in a three-dimensional spatiotemporal coordinate system, the topological fingerprint and the mechanical-metabolic correlation feature set are fused to construct a dynamic intuitionistic fuzzy decision field, and the specific steps are as follows:

[0150] Define the spatial dimensions (x, y) and the time stamp t in a three-dimensional coordinate system;

[0151] Convert the Betti number of the topological fingerprint into a crack membership function;

[0152] Convert the mechanical gradient amplitude into a tissue elasticity non-membership function;

[0153] Fuse the metabolic conflict index to generate a spatiotemporal dynamic hesitation field.

[0154] It should be noted that the spatial dimensions (x, y) and the time stamp t are defined in a three-dimensional coordinate system

[0155] A three-dimensional coordinate system is constructed, in which the spatial dimensions are represented by x and y coordinates to determine the positions of feature points in the tongue image on a plane. At the same time, a time stamp t is introduced to record the state changes of each feature point at different time points. In this way, the static features of the tongue image can be combined with dynamic changes to form a spatiotemporal integrated data framework. In actual operation, continuous monitoring and data collection of the tongue image are required, and the feature data of the tongue image at each time point is marked with the corresponding time stamp, and the position information is recorded in the three-dimensional coordinate system.

[0156] By defining the spatial dimensions and the time stamp in a three-dimensional coordinate system, a spatiotemporal integrated description of the tongue features is achieved. This not only reflects the static features of the tongue at a certain time, but also captures the trend of its changes over time, providing a basis for subsequent dynamic analysis. This spatiotemporal combination method helps to more comprehensively understand the physiological and pathological changes of the tongue, improving the accuracy and reliability of diagnosis.

[0157] Convert the Betti number of the topological fingerprint into a crack membership function

[0158] First, the Betti numbers are extracted from the topological fingerprint. Betti numbers are indices used in topology to describe the number of connected components, loops, and voids in a topological structure. Then, according to the size of the Betti number, it is converted into a crack membership function. This process involves quantitative processing of Betti numbers, converting the complexity of the topological structure into numerical representation of crack membership. Specifically, a mapping relationship can be set to map different Betti number values into the interval of crack membership, thereby obtaining the crack membership function of each feature point.

[0159] Converting the Betti numbers of the topological fingerprint into the crack membership function can convert the complex topological structure information into the membership value that is easy to understand and analyze. This conversion makes the description of crack features more intuitive and quantitative, facilitating subsequent analysis and processing. Through the crack membership function, the existence degree and distribution of cracks in the tongue image can be more clearly identified and evaluated, providing strong support for pathological analysis of tongue images.

[0160] Converting the mechanical gradient amplitude into the tissue elasticity non-membership function

[0161] Obtain the mechanical gradient amplitude data of the tongue image, which reflects the mechanical property changes of the tongue tissue in different regions. Then, process the mechanical gradient amplitude to convert it into a tissue elasticity non-membership function. This process requires establishing a suitable mapping relationship between the size of the mechanical gradient amplitude and the non-membership of tissue elasticity. Specifically, a threshold or function relationship can be set to consider areas with higher gradient amplitude as areas with lower tissue elasticity, thereby obtaining the tissue elasticity non-membership function.

[0162] Converting the mechanical gradient amplitude into the tissue elasticity non-membership function can quantitatively describe the elasticity characteristics of the tongue tissue from a mechanical perspective. This conversion makes the evaluation of tissue elasticity more intuitive and quantitative, facilitating subsequent analysis and processing. Through the tissue elasticity non-membership function, the mechanical property changes of the tongue tissue can be more clearly identified and evaluated, providing strong support for pathological analysis of tongue images.

[0163] Fusion of metabolic conflict index to generate spatiotemporal dynamic hesitation field

[0164] On the basis of the first three steps, the crack membership function, tissue elasticity non-membership function and metabolic conflict index are fused. Metabolic conflict index is an important indicator reflecting the metabolic state of tongue tissue. By combining with crack membership and tissue elasticity non-membership functions, a comprehensive spatiotemporal dynamic hesitation field can be generated. This process involves integration and processing of multi-source data, and a suitable fusion model needs to be established to organically combine different types of feature data, thereby obtaining a hesitation field that can reflect the comprehensive state of tongue image.

[0165] The fusion metabolic conflict index generates a spatiotemporal dynamic hesitation field, which can comprehensively consider the crack characteristics, tissue elasticity and metabolic state of tongue image, and form a comprehensive spatiotemporal dynamic model. This model can more accurately reflect the comprehensive state of tongue image, and provide more rich information for pathological analysis of tongue image. Through the spatiotemporal dynamic hesitation field, the pathological changes of tongue image can be more clearly identified and evaluated, which provides strong support for early diagnosis and treatment of diseases.

[0166] Specifically, the intuitionistic fuzzy decision field is input into a pathological prediction model to output a tongue image pathological evolution risk index and a visual warning report. The specific steps are as follows:

[0167] Discretize the fuzzy decision field into a spatiotemporal grid node;

[0168] Based on Lyapunov index calibration, the chaotic parameter prediction field is evolved;

[0169] Render the pathological risk thermodynamic map in the hyperbolic Poincare disk;

[0170] Integrate the risk gradient to generate an early lesion warning index.

[0171] It should be noted that the fuzzy decision field is discretized into a spatiotemporal grid node

[0172] First, the fuzzy decision field is divided into a plurality of discrete spatiotemporal grid nodes. This process needs to determine the appropriate spatiotemporal resolution, that is, the granularity size of division in time and space dimensions. In the spatial dimension, according to the size and characteristic distribution of tongue image, a suitable grid spacing is selected to divide the tongue image area into a plurality of spatial grid units. In the time dimension, according to the dynamic characteristics of tongue image pathological changes, the time sampling interval is determined to divide the entire time sequence into a plurality of time points. Then, each position and time point in the fuzzy decision field is corresponded to a spatiotemporal grid node to form a discretized spatiotemporal grid model.

[0173] By discretizing the fuzzy decision field into a spatiotemporal grid node, the complex continuous decision field is simplified and quantitatively represented. This discretization process makes the subsequent calculation and analysis more efficient and feasible, and can more accurately capture the pathological change characteristics of tongue image at different spatial positions and time points. At the same time, the discretized spatiotemporal grid model is convenient for fusion and processing with other discretized data, providing a basic data structure for subsequent pathological risk assessment and warning.

[0174] Based on Lyapunov index calibration, the chaotic parameter prediction field is evolved

[0175] Lyapunov exponent is used to calibrate the chaotic parameters to predict the evolution of field strength. First, the data sequence related to field strength is extracted from the discretized spatiotemporal grid nodes. Then, the Lyapunov exponent of the data sequence is calculated, which can reflect the divergence or convergence degree of adjacent trajectories in phase space and is a key indicator to measure the chaotic characteristics of the system. According to the size and trend of Lyapunov exponent, the chaotic parameters are adjusted and calibrated to more accurately reflect the dynamic characteristics of pathological changes in tongue image. Finally, based on the calibrated chaotic parameters, a prediction model is established to predict the evolution of field strength at subsequent time points.

[0176] The prediction of field strength evolution based on Lyapunov exponent calibrated chaotic parameters can effectively capture the complex dynamic characteristics of pathological changes in tongue image. By introducing Lyapunov exponent, the change of chaotic degree of the system can be quantified, so as to more accurately predict the evolution trend of field strength. This prediction method based on chaos theory can better reflect the nonlinearity and uncertainty of pathological changes in tongue image, providing a more reliable basis for early warning of pathological changes.

[0177] Rendering pathological risk thermodynamic map in hyperbolic Poincare disk

[0178] The hyperbolic Poincare disk is selected as the rendering space of the pathological risk thermodynamic map. The hyperbolic Poincare disk is a space model with hyperbolic geometric characteristics, which can represent infinite space in a limited area, suitable for representing complex hierarchical structure and relationship. First, the predicted field strength evolution data is mapped to the coordinate system of the hyperbolic Poincare disk. Then, according to the size and trend of field strength, the pathological risk value of each spatiotemporal grid node is calculated. Finally, the pathological risk value is rendered on the hyperbolic Poincare disk in the form of thermodynamic map, and the risk level is represented by the depth of color.

[0179] Rendering pathological risk thermodynamic map in hyperbolic Poincare disk can more intuitively show the distribution and trend of pathological risk in tongue image. The special geometric characteristics of hyperbolic Poincare disk make the hierarchical structure and mutual relationship of pathological risk more clear, which helps to find potential pathological areas and risk aggregation points. This visualization method provides a intuitive and effective tool for clinicians and researchers to quickly identify and evaluate the pathological risk of tongue image.

[0180] Integrating risk gradient to generate early warning index of pathological changes

[0181] The risk gradient of each spatiotemporal grid node in the pathological risk heat map, that is, the rate of change of the risk value in space and time, is calculated. Then, the risk gradient is integrated to obtain an integrated risk gradient integral value. The integral value reflects the overall trend and intensity of the pathological risk of tongue appearance. Finally, an early lesion warning index is generated according to the size of the integral value, which is used to evaluate the possibility and severity of tongue appearance lesions.

[0182] The integral risk gradient generates an early lesion warning index, which can comprehensively consider the spatiotemporal variation characteristics of the pathological risk of tongue appearance and provide a quantitative and objective warning index. Through integration, local risk gradient information can be integrated into global risk assessment results, thereby more accurately reflecting the overall pathological state of tongue appearance. This warning index provides strong support for the detection and intervention of early lesions and helps to improve the early diagnosis rate and treatment effect of diseases.

[0183] The embodiment also provides a tongue diagnosis feature quantitative analysis system integrating fuzzy theory, which comprises:

[0184] A multi-modal acquisition module synchronously acquires tongue surface Young's modulus distribution data and hyperspectral reflectance data through a micro piezoelectric sensor array and a hyperspectral imaging module integrated in a tongue diagnosis device, and generates a multi-modal tongue appearance fusion data set;

[0185] A crack reconstruction module inputs the hyperspectral reflectance data into a neural network model for crack region segmentation and super-resolution reconstruction, and outputs a crack network graph with geometric topological structure and tongue papilla morphological entropy;

[0186] A topological quantification module performs persistent homology analysis on the crack network graph, calculates crack topological invariants, and generates a topological fingerprint of quantitative crack complexity;

[0187] A cross-scale correlation module identifies a mechanical gradient mutation zone based on the Young's modulus distribution data, extracts a metabolic index in combination with the hyperspectral reflectance, and constructs a mechanical-metabolic correlation feature set;

[0188] A dynamic decision-making module fuses the topological fingerprint and the mechanical-metabolic correlation feature set in a three-dimensional spatiotemporal coordinate system, and constructs a dynamic intuitionistic fuzzy decision field;

[0189] A risk warning module inputs the intuitionistic fuzzy decision field into a pathological prediction model, and outputs a tongue appearance pathological evolution risk index and a visual warning report.

[0190] The embodiment also provides a computer device suitable for the tongue diagnosis feature quantitative analysis method integrating fuzzy theory, which comprises a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the tongue diagnosis feature quantitative analysis method integrating fuzzy theory proposed in the above embodiment.

[0191] The computer device can be a terminal, which includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved by WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, a trackball or a touchpad arranged on the shell of the computer device, or an external keyboard, a touchpad or a mouse, etc.

[0192] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the tongue diagnosis feature quantification analysis method based on fuzzy theory fusion as described in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0193] In summary, the present application realizes the collaborative quantification of the biomechanical properties and optical characteristics of the tongue surface by the multimodal synchronous acquisition of micro-piezoelectric sensors and hyperspectral imaging, breaking through the limitations of traditional tongue diagnosis relying on single image data; generates a topological fingerprint based on the persistent homology analysis of crack networks, converting subjective morphological description into objective topological invariants, significantly improving the quantification accuracy of complex cracks; constructs a cross-scale mechanical-metabolic correlation feature set, establishes the mapping relationship between the elastic modulus gradient of the tongue body and metabolic indicators, and reveals the deep pathological mechanism of the correlation between shape and quality; forms a dynamic intuitionistic fuzzy decision field in a three-dimensional spatiotemporal coordinate system, captures the uncertainty of pathological evolution by fusing the topological fingerprint and mechanical-metabolic features combined with the hesitancy operator, and realizes the leap from static rules to dynamic early warning; finally, through the visualization of hyperbolic space driven by chaos, outputs the early lesion risk thermodynamic map, and provides quantifiable decision support for the traditional Chinese medicine concept of "preventing disease before it occurs".

[0194] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and all should be covered in the scope of the claims of the present application.

Claims

1. A method for quantitative analysis of tongue diagnosis features incorporating fuzzy theory, characterized in that, Includes the following steps: By integrating a micro-piezoelectric sensor array and a hyperspectral imaging module into the tongue diagnosis device, the Young's modulus distribution data and hyperspectral reflectance data of the tongue surface are collected simultaneously to generate a multimodal tongue image fusion dataset. The hyperspectral reflectance data is input into a neural network model for crack region segmentation and super-resolution reconstruction, and the output is a crack network diagram with geometric topology and tongue papilla morphological entropy. A continuous homology analysis is performed on the crack network graph to calculate the crack topological invariants and generate a topological fingerprint that quantifies the crack complexity. Based on the Young's modulus distribution data, mechanical gradient abrupt change regions are identified, and metabolic indices are extracted by combining hyperspectral reflectance to construct a mechanical-metabolic correlation feature set. By fusing the topological fingerprint and the mechanical-metabolic correlation feature set in a three-dimensional spatiotemporal coordinate system, a dynamic intuitionistic fuzzy decision field is constructed. The intuitive fuzzy decision field is input into the pathological prediction model, and the output is a risk index of tongue pathological evolution and a visual early warning report.

2. The method for quantitative analysis of tongue diagnosis features based on fuzzy theory as described in claim 1, characterized in that: The process involves simultaneously acquiring Young's modulus distribution data and hyperspectral reflectance data of the tongue surface using a micro-piezoelectric sensor array and a hyperspectral imaging module integrated into the tongue diagnosis device, generating a multimodal tongue image fusion dataset. The specific steps are as follows: The micro piezoelectric sensor array is controlled to apply gradient pressure in a ring topology arrangement: 0-5 kPa in the central region, 5-10 kPa in the transition region, and 10-15 kPa in the edge region. During the pressure loading process, the hyperspectral imaging module is triggered to capture reflectance data in five bands (540nm, 560nm, 920nm, 970nm, and 620nm). The pressure-deformation curve is calibrated in real time through a dynamic feedback adjustment module to eliminate tongue movement artifacts. The mechanical response data and spectral data are spatiotemporally aligned to generate a fused dataset with three-dimensional coordinate binding.

3. The method for quantitative analysis of tongue diagnosis features based on fuzzy theory as described in claim 2, characterized in that: The hyperspectral reflectance data is input into a neural network model for crack region segmentation and super-resolution reconstruction, outputting a crack network diagram with geometric topological structure and the morphological entropy of the tongue papillae. The specific steps are as follows: Candidate crack regions with saturation > 0.7 and lightness < 0.4 are extracted in the HSV color space; A nonlocal attention U-Net++ model is used to perform sub-pixel-level segmentation of candidate regions and output a binary crack mask. The mask is transformed into a topological graph structure based on the skeletonization algorithm, where the intersection points are vertices and the crack paths are edges. Calculate the curvature histogram of the tongue papilla contours to generate entropy values ​​that characterize morphological complexity.

4. The method for quantitative analysis of tongue diagnosis features based on fuzzy theory as described in claim 3, characterized in that: The crack network graph is subjected to continuous homology analysis to calculate crack topological invariants and generate a topological fingerprint that quantifies crack complexity. The specific steps are as follows: Map the vertex set of the crack topology graph to two-dimensional point cloud data; A scale-increasing topology is constructed using Vietoris-Rips complex filtering; Extract the set of barcodes whose one-dimensional homology groups have a survival interval exceeding 0.15 mm; Calculate the Betti number and persistent entropy of the barcode as a topological fingerprint.

5. The method for quantitative analysis of tongue diagnosis features based on fuzzy theory as described in claim 4, characterized in that: Based on the Young's modulus distribution data, regions of abrupt changes in mechanical gradient are identified. Metabolic indices are extracted using hyperspectral reflectance, and a mechanical-metabolic correlation feature set is constructed. The specific steps are as follows: Anisotropic diffusion filtering is applied to the Young's modulus distribution to calculate the gradient magnitude; Regions with gradient magnitudes > 2.5 kPa / mm are designated as areas of mechanical abrupt change. Oxygenation index and hydration index were extracted in the 540nm and 970nm wavelength bands; Establish a weighted mapping matrix between the mechanical gradient and metabolic index in the mutation region.

6. The method for quantitative analysis of tongue diagnosis features based on fuzzy theory as described in claim 5, characterized in that: In a three-dimensional spatiotemporal coordinate system, the aforementioned topological fingerprint and the mechanical-metabolic correlation feature set are fused to construct a dynamic intuitionistic fuzzy decision field. The specific steps are as follows: Define the spatial dimension (x, y) and the timestamp t in a three-dimensional coordinate system; Transform the Betti number of the topological fingerprint into a crack membership function; The mechanical gradient magnitude is transformed into a tissue elasticity non-membership function; The hesitancy field of spatiotemporal dynamics is generated by the fusion of metabolic conflict index.

7. The method for quantitative analysis of tongue diagnosis features based on fuzzy theory as described in claim 6, characterized in that: The intuitive fuzzy decision field is input into the pathological prediction model, which outputs a tongue pathological evolution risk index and a visual early warning report. The specific steps are as follows: Discretize the fuzzy decision field into spatiotemporal grid nodes; Predicting field strength evolution based on chaotic parameters calibrated by Lyapunov exponent; Rendering a pathological risk heatmap in a hyperbolic Poincaré disk; The integral risk gradient generates an early warning index for lesions.

8. A system for quantitative analysis of tongue diagnosis features based on fuzzy theory, and a method for quantitative analysis of tongue diagnosis features based on fuzzy theory as described in any one of claims 1 to 7, characterized in that: include, The multimodal acquisition module, through the micro piezoelectric sensor array and hyperspectral imaging module integrated into the tongue diagnosis device, simultaneously acquires Young's modulus distribution data and hyperspectral reflectance data of the tongue surface, generating a multimodal tongue image fusion dataset; The crack reconstruction module inputs the hyperspectral reflectance data into a neural network model to perform crack region segmentation and super-resolution reconstruction, and outputs a crack network diagram with geometric topology and tongue papilla morphological entropy. The topology quantization module performs continuous homology analysis on the crack network graph, calculates the crack topology invariants, and generates a topology fingerprint that quantifies the crack complexity. The cross-scale correlation module identifies abrupt changes in mechanical gradients based on the Young's modulus distribution data, extracts metabolic indices by combining hyperspectral reflectance, and constructs a mechanical-metabolic correlation feature set. The dynamic decision-making module integrates the topological fingerprint and the mechanical-metabolic correlation feature set in a three-dimensional spatiotemporal coordinate system to construct a dynamic intuitionistic fuzzy decision field; The risk warning module inputs the intuitive fuzzy decision field into the pathological prediction model and outputs a risk index of tongue pathological evolution and a visual warning report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the tongue diagnosis feature quantification analysis method based on fuzzy theory as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the tongue diagnosis feature quantification analysis method based on the fusion fuzzy theory described in any one of claims 1 to 7.

Images