Real-time noninvasive corneal glucose detection system and method thereof

Abstract

The invention discloses a corneal glucose real-time non-invasive detection system and method, and belongs to the technical field of glucose real-time non-invasive detection.The method comprises the following steps that corneal reflection spectrum, tear film thickness change and environment temperature and humidity data are obtained in real time, and spectrum compensation parameters are generated through a temperature and humidity compensation model; outputting a tear film refractive index correction value in combination with a refractive index correction model, and correcting the reflection spectrum through a combined correction formula; then, calculating a reference reflection spectrum and a deviation value based on historical data, extracting statistical characteristics, and fusing confidence to obtain a confidence factor for adjusting the deviation value; if the deviation exceeds the threshold value, replacing the corrected data with the reference data; and finally, calculating the glucose concentration according to the processed spectrum, and determining the health grade according to a preset comparison table. The high-reliability non-invasive blood glucose health monitoring closed-loop method and system are constructed through multi-source data fusion correction, intelligent deviation optimization and a confidence coefficient decision mechanism.

Description

TECHNICAL FIELD

[0001] The application relates to a real-time non-invasive detection technology field of glucose, in particular to a corneal glucose real-time non-invasive detection system and a method thereof. BACKGROUND

[0002] With the rising incidence of metabolic diseases such as diabetes, non-invasive blood glucose detection technology has attracted widespread attention in the field of health monitoring due to its simple operation and avoidance of infection risks. Among them, the detection method based on corneal or tear glucose concentration as a potential non-invasive means has been preliminarily explored in recent years. The corneal glucose concentration is correlated with the blood glucose level, and the data is obtained by non-contact means such as optical reflection spectrum, which has the advantages of real-time and painless, and is suitable for daily health management.

[0003] Most of the existing corneal glucose detection schemes are based on a single optical parameter or simple spectral analysis. The typical approach is to use a spectrometer to directly collect corneal reflection spectrum data and directly calculate the glucose concentration value through a pre-set calibration curve. The advantage of such prior art is that the device structure is relatively simple, the measurement process is fast, and it can reflect the trend of glucose concentration to a certain extent.

[0004] However, the prior art has the following limitations. First, its detection mechanism is "static", lacking comprehensive consideration of environmental influences (such as temperature and humidity changes) and physiological states (such as fluctuations in tear film thickness). The refractive index of the tear film is affected by the thickness and environmental temperature and humidity, and the existing method does not introduce an effective correction model, resulting in distorted spectral data and reduced measurement accuracy. Secondly, even if a correction model is introduced, the existing technology lacks an intelligent decision-making core that can learn from historical data and dynamically optimize. This results in large environmental interference and poor stability of the detection results, making it difficult to cope with individual differences or complex use scenarios (such as different climate conditions and user tear film characteristics). Therefore, we propose a real-time non-invasive detection method for corneal glucose to solve this problem. SUMMARY

[0005] In view of the defects of the existing emerging optical non-invasive blood glucose detection technology, such as insufficient compensation for environmental interference and uncorrected tear film noise, it is difficult to meet the clinical precise monitoring requirements.

[0006] In order to achieve the above purpose, the technical scheme of the present application is as follows:

[0007] In a first aspect, the application discloses a real-time non-invasive detection method for corneal glucose, comprising the following steps:

[0008] A data acquisition module is used to acquire corneal reflection spectrum data, tear film thickness change data and environmental temperature and humidity data of a target object in real time.

[0009] a data processing module, configured to input the temperature and humidity data into a pre-constructed temperature and humidity compensation model to generate a spectrum compensation parameter and a parameter confidence level, and to input the temperature and humidity data and the tear film thickness change data into a pre-constructed refractive index correction model to generate a tear film refractive index correction amount and a correction amount confidence level;

[0010] a joint correction processing module, configured to perform joint correction on the reflected spectrum data by using the spectrum compensation parameter and the tear film refractive index correction amount through a joint correction formula; the joint correction formula is: wherein, is real-time acquired reflected spectrum data, is the spectrum compensation parameter, is the tear film refractive index correction amount, is a reference refractive index of the tear film;

[0011] a spectrum effectiveness verification module, configured to calculate reference reflected spectrum data based on a database constructed based on historical reflected spectrum data, to calculate a deviation value set of the corrected reflected spectrum data and the reference reflected spectrum data by using the associated tear film thickness change data and the temperature and humidity data, to extract statistical distribution characteristics of the deviation value set as a deviation value, to perform weighted fusion on the parameter confidence level and the correction amount confidence level to obtain a confidence factor, and to adjust the deviation value based on the confidence factor, and if the deviation value is greater than a preset deviation threshold, to replace the corrected reflected spectrum data with the reference reflected spectrum data;

[0012] a data analysis module, configured to calculate a corneal glucose concentration value based on the processed reflected spectrum data, and to obtain a health level of the target object by referring to a preset control table representing a mapping relationship between the corneal glucose concentration and the health level.

[0013] In a second aspect, the present application introduces a corneal glucose real-time non-invasive detection system, comprising the following modules:

[0014] a data acquisition module, configured to acquire, in real time, corneal reflected spectrum data, tear film thickness change data of a target object, and temperature and humidity data of an environment in which the target object is located;

[0015] a data processing module, configured to input the temperature and humidity data into a pre-constructed temperature and humidity compensation model to generate a spectrum compensation parameter and a parameter confidence level, and to input the temperature and humidity data and the tear film thickness change data into a pre-constructed refractive index correction model to generate a tear film refractive index correction amount and a correction amount confidence level;

[0016] a joint correction processing module, configured to perform joint correction on the reflected spectrum data by using the spectrum compensation parameter and the tear film refractive index correction amount through a joint correction formula; the joint correction formula is: wherein, is a real-time acquired reflection spectrum data, is a spectrum compensation parameter, is a tear film refractive index correction amount, is a tear film reference refractive index;

[0017] a spectrum validity verification module, configured to calculate reference reflection spectrum data based on a database constructed based on historical reflection spectrum data, associated tear film thickness change data, temperature and humidity data, calculate a deviation value set of the corrected reflection spectrum data and the reference reflection spectrum data, and extract statistical distribution characteristics of the deviation value set as a deviation value; further configured to weight and fuse the parameter confidence and the correction amount confidence to obtain a confidence factor; further configured to adjust the deviation value based on the confidence factor, and if the deviation value is greater than a preset deviation threshold, replace the corrected reflection spectrum data with the reference reflection spectrum data;

[0018] a data analysis module, configured to calculate a corneal glucose concentration value according to the processed reflection spectrum data, and then refer to a preset control table representing a mapping relationship between the corneal glucose concentration and the health degree to obtain a health level of the target object.

[0019] In a third aspect, the present application provides a computing device, comprising:

[0020] a memory, configured to store a program;

[0021] a processor, configured to execute the computer executable instruction, and the computer executable instruction realizes the steps of the corneal glucose real-time non-invasive detection method when executed by the processor.

[0022] In a fourth aspect, the present application provides a computer readable storage medium, comprising: the program is executed by the processor, and the steps of the corneal glucose real-time non-invasive detection method are realized.

[0023] Compared with the prior art, the present application has the following beneficial effects:

[0024] The present application synchronously acquires corneal reflection spectrum, tear film thickness and environmental temperature and humidity data through a multi-source data synchronous acquisition and joint correction mechanism, generates a composite correction parameter based on a temperature and humidity compensation model and a refractive index correction model, and realizes multi-dimensional compensation of the reflection spectrum through a joint correction formula, effectively solving the coupling influence of environmental interference and tear film transients on optical measurement, and improving the quality and stability of the spectrum data.

[0025] The application constructs a benchmark spectrum library based on historical data, extracts statistical deviation values through multi-dimensional feature comparison of real-time spectrum and benchmark spectrum, and generates a dynamic adjustment factor by fusing the confidence of the spectrum compensation parameter and the refractive index correction amount, realizes intelligent verification and self-adaptive optimization of the corrected spectrum data, solves the problem of insufficient data reliability caused by physiological fluctuations and environmental changes in traditional optical detection, and significantly reduces the risk of misjudgment.

[0026] The application realizes the automatic benchmark spectrum replacement strategy when the data quality is poor through the confidence weighted deviation threshold judgment mechanism, ensures the data basis reliability of subsequent glucose concentration inversion, and realizes the closed-loop quality control from original spectrum collection to final health evaluation based on the mapping relationship between glucose concentration and health level, which provides a complete technical scheme with accuracy, robustness and practicality for non-invasive blood glucose monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0027] The disclosure of the application will be described with reference to the accompanying drawings. It should be understood that the drawings are only for illustrative purposes, and are not intended to limit the scope of protection of the application. In the drawings, the same reference numerals are used to refer to the same parts. Among them:

[0028] Figure 1 The flowchart of the corneal glucose real-time non-invasive detection method introduced by the application;

[0029] Figure 2 The flowchart of the construction of the temperature and humidity compensation model based on Figure 1 ;

[0030] Figure 3 The flowchart of the construction of the tear film refractive index correction model based on Figure 1 ;

[0031] Figure 4 The flowchart of the construction of the glucose concentration prediction model. DETAILED DESCRIPTION

[0032] It is easy to understand that according to the technical scheme of the application, those skilled in the art can propose a plurality of structure modes and implementation modes which can be replaced with each other without changing the essential spirit of the application. Therefore, the following specific embodiments and drawings are only exemplary description of the technical scheme of the application, and should not be regarded as the whole or as the limitation or restriction of the technical scheme of the application.

[0033] In the prior art, the emerging optical non-invasive blood glucose detection technology has the problems of insufficient environmental interference compensation, uncorrected tear film noise, etc., which is difficult to meet the clinical precise monitoring demand.

[0034] In order to solve the above problems, as Figure 1As shown, the present application proposes a real-time non-invasive detection method for corneal glucose, comprising the following steps:

[0035] S1. A data acquisition module for acquiring corneal reflectance spectrum data, tear film thickness change data of a target object and temperature and humidity data of the environment where the target object is located in real time;

[0036] S2. Input the temperature and humidity data into the pre-constructed temperature and humidity compensation model to generate spectrum compensation parameters and parameter confidence; also used for inputting the temperature and humidity data and tear film thickness change data into the pre-constructed refractive index correction model to generate tear film refractive index correction amount and correction amount confidence;

[0037] S3. The spectrum compensation parameters and tear film refractive index correction amount are corrected by a joint correction formula to the reflectance spectrum data; the joint correction formula is: Wherein, is the real-time acquired reflectance spectrum data, is the spectrum compensation parameter, is the tear film refractive index correction amount, is the tear film reference refractive index;

[0038] S4. Based on the database constructed by historical reflectance spectrum data, calculate the reference reflectance spectrum data through the associated tear film thickness change data and temperature and humidity data, calculate the deviation value set of the corrected reflectance spectrum data and the reference reflectance spectrum data, and extract the statistical distribution characteristics of the deviation value set as the deviation value;

[0039] S5. Also used for weighting and fusing the parameter confidence and the correction amount confidence to obtain a confidence factor; also used for adjusting the deviation value based on the confidence factor, if the deviation value is greater than a preset deviation threshold, the reference reflectance spectrum data is used to replace the corrected reflectance spectrum data;

[0040] S6. Calculate the corneal glucose concentration value according to the processed reflectance spectrum data, and then refer to a preset control table representing the mapping relationship between corneal glucose concentration and health degree to obtain the health level of the target object.

[0041] After introducing the basic idea of the present application, the embodiments of the present application will be specifically introduced with reference to the accompanying drawings.

[0042] Regarding step S1:

[0043] The process of data acquisition includes:

[0044] (1) Data acquisition stage

[0045] The corneal reflectance spectrum data of the target object is collected in real time by the optical sensing unit;

[0046] synchronously acquiring tear film thickness change data of the target object through a biological measurement module;

[0047] synchronously collecting temperature and humidity data of an environment where the target object is located through an environment monitoring unit;

[0048] (2) Data processing stage

[0049] Signal preprocessing is performed on the corneal reflection spectrum data, including filtering and denoising and signal amplification.

[0050] The tear film thickness change data is processed for time alignment to keep it synchronized with the reflection spectrum data.

[0051] The temperature and humidity data are subjected to analog-to-digital conversion and standardization processing.

[0052] (3) Data output stage

[0053] The processed corneal reflection spectrum data, tear film thickness change data and temperature and humidity data are output to a subsequent data correction module for joint correction processing.

[0054] The optical sensing unit includes a spectrometer and a matching optical probe, and the optical probe is configured to emit probe light to the cornea at a specific incident angle and receive reflected light signals.

[0055] The biological measurement module uses any one of optical coherence tomography technology or laser interferometry technology.

[0056] The environment monitoring unit includes a temperature sensor and a humidity sensor, which are integrated on the probe housing of the optical sensing unit or arranged as independent modules in the measurement environment.

[0057] The signal preprocessing further includes baseline correction and spectrum range interception of the corneal reflection spectrum data.

[0058] Through the above data acquisition scheme, the present application realizes multi-source synchronous acquisition and collaborative processing of the corneal reflection spectrum of the target object, the tear film thickness change and the environmental temperature and humidity. Through parallel implementation of optical sensing and biological measurement, the strict correspondence of key physiological parameters and environmental parameters in time sequence is ensured, which provides a complete and synchronous data basis for subsequent joint correction. The configuration of the special optical probe ensures the acquisition quality of the corneal reflection spectrum, and the selection of multiple biological measurement technologies enhances the adaptability and reliability of the system. Through the standardized preprocessing of the data of each channel, not only the noise interference in the signal acquisition process is eliminated, but also a unified data specification is established, which creates necessary conditions for subsequent data fusion and accurate analysis. Through the collaborative work of multiple modules, the scheme constructs a complete and reliable front-end data acquisition and processing pipeline, which provides key data guarantee for the accurate implementation of non-invasive blood glucose detection.

[0059] Regarding step S2:

[0060] The role of the temperature and humidity compensation model is to eliminate the interference of environmental temperature and humidity changes on spectral detection, and to ensure the accuracy of glucose concentration detection under different environmental conditions. After compensation by the model, the system can still maintain high precision under extreme temperature and humidity environments (such as low temperature and low humidity or high temperature and high humidity), avoiding detection errors caused by environmental changes.

[0061] As shown in Figure 2 The establishment process of the temperature and humidity compensation model is divided into the following three stages:

[0062] (1) Data acquisition stage

[0063] A PID controller is used to accurately adjust the temperature and humidity of the environment box (temperature range: 15-35℃, humidity range: 20%-80%), and the temperature and humidity combinations are changed in steps (such as 5℃×5% grid) and recorded.

[0064] The light source can emit multiple waveband near-infrared light, such as near-infrared light with wavelengths of 1050nm and 1300nm, which irradiates standard glucose solutions with different concentration gradients (such as 0-200mg / dL, with an interval of 20mg / dL).

[0065] Each concentration is repeated 10 times under the same temperature and humidity conditions, and the reflected spectrum is collected by a photodetector (resolution 0.5nm).

[0066] High-precision environmental sensor data (temperature accuracy ±0.1℃, humidity accuracy ±1%) are recorded synchronously, and all devices are time-synchronized through PTP protocol (deviation <200μs).

[0067] The reference temperature is set to 25 DEG C and the reference humidity is set to 40%, and the obtained environmental temperature and humidity data are processed to obtain temperature offset and humidity offset. A four-dimensional data set is constructed according to the obtained spectral data, glucose concentration, temperature offset and humidity offset. The data set covers extreme environmental scenarios: high temperature and high humidity (35 DEG C, 80%), low temperature and low humidity (15 DEG C, 20%) and the like boundary conditions.

[0068] (2) Data processing stage

[0069] The original data is converted into a temperature and humidity compensation model parameter, and a temperature and humidity compensation model is obtained.

[0070] Step 1: data preprocessing

[0071] The reflectivity of each wavelength point is processed by removing outliers using the IQR (Interquartile Range) method. Each wavelength point after removing outliers is independently standardized by Zscore.

[0072] The data is augmented by adding Gaussian noise with a specified signal-to-noise ratio SNR=40dB and expanding the data set by three-dimensional linear interpolation (concentration x temperature x humidity) between adjacent temperature and humidity points.

[0073] Step 2: feature engineering

[0074] The preprocessed data set is filtered for wavelength using a random forest algorithm to obtain processed spectral features, and interference features are extracted.

[0075] Step 3: model training

[0076] The wavelength is weighted to reduce the mean square error (MSE), and the data set is optimized and trained by a full convolutional network (FCN).

[0077] Step 4: model validation

[0078] The stability of the model is evaluated by multiple rounds of cross-validation of the key indicators (wavelength average absolute error) error values.

[0079] (3) Data output stage

[0080] The mathematical representation of the model is obtained by symbolic regression combined with physical constraints to ensure physical reasonableness (such as alpha and beta being negative), and the coefficient accuracy is improved by linear compensation and constraint optimization.

[0081] Temperature and humidity compensation model:

[0082] .

[0083] wherein, Temperature and humidity compensation quantity (reflectance change quantity), Current ambient temperature (℃), Current ambient humidity (%), Reference temperature (25℃), Reference humidity (40%), Temperature compensation coefficient (unit: 1 / ℃), Humidity compensation coefficient (unit: 1 / %), Temperature and humidity cross compensation coefficient (unit: 1 / (%·℃)), , , Model coefficient, obtained by processing data, and , , .

[0084] The model itself has no dynamic adjustment threshold, but when using the compensation quantity, the system sets the compensation quantity boundary (|ΔR|≤0.1). When the calculated value is out of range, the following operations are triggered:

[0085] If |ΔR|>0.1, process according to 0.1, and issue an environmental abnormality warning.

[0086] If |ΔR|<0.1, process according to 0.1, and issue an environmental abnormality warning.

[0087] The role of the tear film refractive index correction model is to eliminate the interference of tear film thickness changes on spectral detection, and to ensure the accuracy of glucose concentration detection under different tear film thickness conditions. After correction by the model, the system can still maintain high precision when the tear film thickness changes, avoiding detection errors caused by changes in tear film thickness.

[0088] As shown in Figure 3 , the tear film refractive index correction model acquisition method includes the following steps:

[0089] (1) Data acquisition stage

[0090] The low coherence interferometer is used to collect tear film thickness data in real time, and the sampling frequency is not less than 200Hz. At the same time, the environmental temperature (℃) and humidity (%) collected by the environmental sensor are recorded in real time. All data are time-stamped synchronized by PTP protocol, and the time deviation is less than 200μs.

[0091] Through the collected tear film thickness data, environmental temperature (℃) and humidity ​​​The data is used to build a multi-modal data set covering three groups of subjects, namely healthy people, diabetic patients and dry eye patients, each group having a sample size of no less than 100.

[0092] (2) Data processing phase

[0093] Step 2.1: Data preprocessing

[0094] The IQR method is used to remove outliers in the thickness data of the constructed multi-modal data set, with a constraint range of 2.02.5 μm. The effective range of the data is calculated and determined by the IQR method

[0095] The mean of the overall data, the standard deviation of the overall data and the observation value of the individual are calculated. The refractive index data is standardized by Zscore, which converts different orders of magnitude data into the same order of magnitude, and uses the calculated Zscore value to measure, so as to ensure the comparability between data.

[0096] The sampling density is improved by 300% by adding Gaussian noise with a mean of 0 and a standard deviation of 0.05 to the thickness change rate and inserting linear time series interpolation data between adjacent sampling points, thereby enhancing the data.

[0097] Step 2.2: Feature engineering

[0098] The key features of thickness change rate, temperature offset and humidity offset are extracted, and the high-dimensional data samples are projected into the best discriminant vector space to ensure the best separability of the sample data in this space.

[0099] In the original training set after extracting the key features, Bootstrap (random sampling with replacement) method is used to extract multiple training subsets.

[0100] In the training process of each decision tree, the optimal feature is selected for splitting in a randomly selected feature subset.

[0101] Using the above extracted samples and feature subsets, multiple decision trees are trained. Each tree grows as much as possible without pruning.

[0102] Through voting (classification problem) or averaging (regression problem), the prediction results of all decision trees are integrated, the feature importance is selected based on the random forest algorithm, the features of each sample object are ranked in descending order according to the weight, and the top three features are retained to obtain the optimized multi-modal data set,

[0103] The core of the random forest algorithm is that each decision tree is trained on a random subset of the data set, and when selecting a split attribute, a random subset of the attributes is also selected. This reduces the correlation between trees and improves the generalization ability of the model.

[0104] Step 2.3: Model Training

[0105] The processed data is divided into training, validation, and test sets for model training, hyperparameter tuning, and performance evaluation, respectively.

[0106] The structure of the neural network is constructed, including input layer, hidden layer, and output layer. The PyTorch framework is used to define the number of model neurons and activation functions.

[0107] The input layer receives the original training set data without calculation and only passes it to the next layer.

[0108] The hidden layer uses a weighted MSE loss function, with the sample weight of diabetic patients increased to 1.5 times, to measure the error between the predicted value and the true value. The gradient of each parameter is calculated by the loss function, and the error is propagated from the output layer to the input layer using the chain rule, extracting and combining the abstract features of the training set data layer by layer.

[0109] The optimizer uses the stochastic gradient descent (SGD) algorithm to update the model parameters based on the gradient calculated by backpropagation, accelerating convergence and improving training stability.

[0110] The output layer outputs the final prediction based on the task type.

[0111] The input data is passed through the network layers (input layer, hidden layer, and output layer) layer by layer, processed by weight calculation and activation function, and finally generates the prediction result in the output layer.

[0112] The model is trained through multiple iterations (epochs), each iteration uses the training data for forward propagation, loss calculation, and parameter update. At the same time, the validation set is used to monitor the model performance to avoid overfitting.

[0113] After training is completed, the test set is used to evaluate the generalization ability of the model. By calculating accuracy, loss value, and other indicators, the performance of the model on new data is judged.

[0114] (3) Data Output Stage

[0115] Generate a physically interpretable formula through symbolic regression, output the tear film refractive index correction model mathematical formula;

[0116] The process of generating a physically interpretable formula through symbolic regression is to combine data-driven methods with physical prior knowledge to produce mathematical expressions that not only conform to data rules but also have physical explanations. The following is a detailed step-by-step description:

[0117] Step 3.1 Problem Definition and Data Preparation

[0118] Determine input variables: tear film thickness change rate, ambient temperature ( ), ambient humidity ( ), etc.

[0119] Determine output variable: tear film refractive index correction formula: .

[0120] Collect high-quality dataset, including input and output data under different conditions (healthy population, diabetic patients, etc.).

[0121] Step 3.2 Set physical constraints

[0122] According to biomedical knowledge, set the conditions that the formula must satisfy. For example:

[0123] Since tear film evaporation will cause the refractive index to rise, the coefficient of thickness change rate (negative value indicating evaporation) should be negative.

[0124] Temperature rise will accelerate evaporation, so the temperature term coefficient should be negative.

[0125] Humidity increase will slow down evaporation, so the humidity term coefficient should be positive.

[0126] The formula does not contain operations such as division that may produce singular points.

[0127] The complexity (depth) of the formula does not exceed 3 layers to ensure simplicity.

[0128] Step 3.3 Symbolic regression execution

[0129] Initialization: Randomly generate a set of initial formulas (population), each formula consisting of basic operators (add, subtract, multiply) and operands (input variables, constants) in tree structure.

[0130] Evaluation: Calculate the mean square error (MSE) of each formula on the dataset, and calculate the fitness (fitness) combined with the degree of satisfaction of physical constraints. For example: fitness=1 / (1+MSE) constraint satisfaction factor, where the constraint satisfaction factor is 1 if the constraint is met, otherwise 0.1 (penalty).

[0131] Selection: Select excellent individuals according to fitness to enter the next generation (the higher the fitness, the greater the probability of being selected).

[0132] Crossover: Randomly select two parent formulas and exchange their sub-trees to generate new formulas.

[0133] Mutation: Randomly modify part of the formula, including: replace operator, replace sub-tree, perturb constant, etc.

[0134] Iteration: Repeat the evaluation of the selected crossover mutation process until the maximum number of iterations is reached or a formula that meets the accuracy requirement is found.

[0135] Step 3.4 Formula Refinement

[0136] Numerically optimize (e.g., gradient descent) the optimal formula obtained from symbolic regression to fine-tune its constant coefficients for further accuracy improvement.

[0137] Add a regularization term (e.g., L2 regularization) during optimization to prevent overfitting and ensure that the coefficients are within a reasonable range.

[0138] Step 3.5 Physical Verification

[0139] Verify that the final formula meets all pre-set physical constraints.

[0140] Evaluate the interpretability of the formula through clinical data.

[0141] Verify the predictive performance of the formula on an independent test set.

[0142] Step 3.6 Output Formula

[0143] Express the formula as a mathematical expression, e.g.:

[0144]

[0145] Record the physical meaning and numerical value of each coefficient in the formula.

[0146] Example:

[0147] Suppose, through the above process, the formula is obtained:

[0148]

[0149] Where:

[0150] The first term represents the dominant effect of evaporation (the faster the thickness decreases, the more the refractive index increases);

[0151] The second term represents the effect of temperature increase on evaporation (thus increasing the refractive index);

[0152] The third term represents the effect of humidity increase on evaporation (thus decreasing the refractive index).

[0153] Express the tear film refractive index correction model as a mathematical expression:

[0154]

[0155] Where, is the refractive index correction amount, is the reference refractive index, taken as 1.336, For the change in refractive index.

[0156] The tear film refractive index correction amount is obtained by inputting the tear film change data into the tear film refractive index correction model.

[0157] The joint correction is a key step of combining the spectral compensation parameter and the tear film refractive index correction amount. The joint correction can comprehensively eliminate the effects of environmental temperature and humidity changes, changes in refractive index caused by tear film evaporation, and light intensity on spectral data, and provide accurate spectral correction data for glucose concentration calculation. Through multi-source data fusion, more robust correction effect is achieved.

[0158] Regarding step S3:

[0159] The joint correction formula generation process includes the following steps:

[0160] (1) Data acquisition stage

[0161] The spectral compensation amount ΔS is obtained, which is calculated by the environmental temperature T and humidity H collected by the temperature and humidity sensor in real time through the environmental temperature and humidity compensation model.

[0162] The tear film refractive index correction amount is obtained, which is calculated by combining the environmental data through the tear film refractive index correction model by measuring the tear film thickness change rate by a low coherence interferometer.

[0163] These data are time-aligned under a time stamp synchronization mechanism (PTP protocol) with a deviation of <200μs.

[0164] (2) Data processing stage

[0165] In the data processing stage, the system performs the following steps to generate the joint correction formula:

[0166] Step 2.1: Physical effect modeling

[0167] Optical path correction: The change in optical path length caused by the change in tear film refractive index, according to the Fresnel reflection law, the relationship between the reflected light intensity and the refractive index is:

[0168] Therefore, the effect of refractive index change on the reflected spectrum is corrected by the factor , where is the reference refractive index (1.336).

[0169] Environmental interference correction: The spectral baseline drift caused by temperature and humidity changes is corrected by the spectral compensation parameter ΔS.

[0170] Device drift correction: The gain change caused by device aging or state change is corrected by the device correction coefficient.

[0171] Step 2.3: Formula integration

[0172] Integrate the above correction terms, generate a physically interpretable formula through symbolic regression, and form a joint correction formula from the physically interpretable formula:

[0173]

[0174] wherein, wherein: is the original spectral reflectance intensity, is the spectral compensation parameter (obtained from the temperature and humidity compensation model), is the corrected tear film refractive index (obtained from the tear film refractive index correction formula), is the tear film reference refractive index (1.336), is the device drift correction coefficient.

[0175] (3) Data output stage

[0176] In the data output stage, the system outputs the corrected spectrum for glucose concentration calculation.

[0177] The joint correction formula simultaneously considers the interference of environment, physiology and device in three dimensions, integrates three independent compensation terms into one formula, and realizes it through hard constraint to ensure that the corrected spectrum will not exceed the physical reasonable range, avoiding false compensation.

[0178] Regarding step S4:

[0179] Spectrum data verification based on confidence includes:

[0180] Obtain the real-time reflectance spectrum data generated after joint correction;

[0181] Call the pre-constructed historical reflectance spectrum database, which stores a plurality of reference reflectance spectrum data collected under controlled conditions and verified;

[0182] Reference comparison and deviation extraction: compare the real-time reflectance spectrum data with the corresponding reference reflectance spectrum data retrieved from the historical database, calculate the difference between the two in multi-dimensional features, form a deviation value set, and extract the statistical distribution characteristics representing the overall deviation degree from the set as a comprehensive deviation value;

[0183] Wherein, the statistical distribution characteristics include but are not limited to at least one of the mean, standard deviation, or specific quantile of the deviation value set.

[0184] The method for constructing a reference reflectance spectrum database based on historical reflectance spectrum data comprises the following steps:

[0185] (1) Data acquisition stage

[0186] collecting and storing a plurality of sets of historical reflectance spectrum data, wherein each set of historical reflectance spectrum data is associated with environmental parameters, tear film state parameters and corresponding verified glucose concentration values at the time of collection;

[0187] the environmental parameters at least include environmental temperature and humidity data, and the tear film state parameters at least include tear film thickness variation data;

[0188] (2) data processing stage

[0189] data cleaning and screening are performed on the historical reflectance spectrum data, and effective historical reflectance spectrum data meeting preset quality standards are retained;

[0190] the effective historical reflectance spectrum data are subjected to cluster analysis according to their associated environmental parameters, tear film state parameters and glucose concentration values, to form a plurality of data clusters with similar characteristics;

[0191] for the reflectance spectrum data in each data cluster, reference reflectance spectrum data corresponding to the data cluster are generated by calculating the central tendency of the spectral characteristics of the reflectance spectrum data;

[0192] (3) data output stage

[0193] all data clusters and their corresponding reference reflectance spectrum data are output and stored, to construct the reference reflectance spectrum database;

[0194] each reference reflectance spectrum data is used as a standardized spectral reference for comparison under specific environmental conditions and tear film states.

[0195] wherein data cleaning and screening include removing spectral data with a signal-to-noise ratio below a threshold value, signal intensity beyond a reasonable range or with obvious motion artifacts.

[0196] wherein the reference reflectance spectrum data are generated by calculating the central tendency of the spectral characteristics, specifically by calculating the arithmetic mean or median value of all reflectance spectrum data in the data cluster at each wavelength point.

[0197] wherein, for each data cluster and its reference reflectance spectrum data, an index is established, which is constructed based at least on the corresponding environmental parameter range, tear film state parameter range and glucose concentration range.

[0198] Regarding step S5:

[0199] confidence-based spectral data replacement includes:

[0200] A confidence factor fusion is performed on the spectral compensation parameter and the tear film refractive index correction amount used in the joint correction process, and data quality confidence of the two is respectively evaluated, and the two confidence degrees are fused according to a predefined weight ratio to generate a comprehensive confidence factor.

[0201] An adaptive bias adjustment and judgment is performed on the comprehensive bias value by using the confidence factor, wherein the confidence factor is used to reflect the reliability of the current correction process, and the adjusted comprehensive bias value is compared with a preset bias threshold value.

[0202] When the adjusted comprehensive bias value is greater than the preset bias threshold value, it is determined that the real-time corrected reflectance spectrum data is unreliable, and a control instruction is output to replace the real-time corrected reflectance spectrum data with the reference reflectance spectrum data retrieved from the historical database for subsequent glucose concentration inversion calculation.

[0203] The confidence evaluation of the spectral compensation parameter and the tear film refractive index correction amount is based on at least one of the noise level of the respective data source, the model fitting residual, or the degree of conformity with the historical normal value range.

[0204] The dynamic adjustment of the comprehensive bias value by using the confidence factor is specifically: when the confidence factor is lower than a set level, the comprehensive bias value is correspondingly increased to more easily trigger data replacement.

[0205] The preset bias threshold value is preset based on the long-term statistical characteristics of the verified data in the historical reflectance spectrum database, and can be periodically optimized with the update of the database.

[0206] Through the above technical solutions, the application constructs a complete spectrum data quality guarantee system, and realizes intelligent verification and adaptive optimization of real-time measurement data. By establishing a reference spectrum library based on historical data, a reliable comparison standard is provided for real-time data quality evaluation. Through the confidence fusion mechanism, the reliability of environmental compensation and refractive index correction is quantified into the data verification link, which improves the accuracy of abnormal data identification. The scheme adopts a combination strategy of multi-dimensional feature comparison and dynamic threshold adjustment, which not only ensures the scientificity of data replacement decision, but also ensures the adaptability of the system under different measurement conditions. In addition, the reference data generation method based on clustering analysis enables the system to automatically construct reference standards covering multiple physiological states and environmental conditions, enhancing the robustness and clinical applicability of the non-invasive blood glucose detection system. Through this closed-loop data quality management mechanism, the application effectively solves the data reliability problem caused by environmental interference and physiological changes in optical non-invasive detection, and provides key technical support for precise blood glucose monitoring.

[0207] Regarding step S6:

[0208] The corrected reflectance spectrum data is used to calculate the corneal glucose concentration value, specifically as follows:

[0209] The corrected reflectance spectrum data is input into the glucose concentration prediction model to generate the corneal glucose concentration value.

[0210] As shown in Figure 4 , the method for constructing the glucose concentration prediction model includes the following steps:

[0211] In a controlled environment, a plurality of sets of corneal reflectance spectrum data corresponding to known blood glucose concentrations are obtained;

[0212] The spectral data is subjected to noise reduction filtering, baseline correction, and feature wavelength extraction processing.

[0213] The spectral data is converted into a digital signal, and a spectral feature vector is extracted.

[0214] A mapping relationship model between the spectral feature vector and the blood glucose concentration is trained and constructed by a partial least squares regression algorithm, the processed spectral feature vector is input into an initial model and iteratively optimized to obtain the temperature and humidity compensation model.

[0215] The corrected reflectance spectrum data is used to calculate the corneal glucose concentration value.

[0216] The method for constructing a mapping relationship model between a spectral feature vector and a blood glucose concentration based on a partial least squares regression includes the following steps:

[0217] (1) Data acquisition stage

[0218] Near-infrared or Raman spectrum data of biological samples are collected as a set of independent variables, and blood glucose concentration reference values obtained by clinical testing are collected as a set of dependent variables; a training data set is constructed from the above data

[0219] (2) Data processing stage

[0220] Step 2.1

[0221] The spectral data is subjected to centering processing to eliminate the influence of spectral baseline offset, standardization processing to unify the data scale of each wavelength point, and convolution smoothing filtering to suppress high-frequency noise to realize data enhancement processing

[0222] Step 2.2

[0223] The prediction error corresponding to different latent variable numbers is calculated by cross-validation, and the latent variable number that minimizes the prediction error is selected as the model complexity parameter.

[0224] Step 2.3

[0225] Perform partial least squares regression modeling:

[0226] Initialize the spectral data residual matrix and the blood glucose concentration residual vector;

[0227] Iteratively calculate each latent variable component in order:

[0228] a. Calculate the spectral weight vector, which maximizes the covariance of the spectral residual and the blood glucose concentration residual in direction;

[0229] b. Calculate the sample projection score vector, which represents the distribution of the sample in the latent variable space;

[0230] c. Calculate the spectral loading vector and the blood glucose concentration weight coefficient, which respectively describe the association strength of the latent variable and the original spectrum, blood glucose concentration;

[0231] d. Update the spectral and blood glucose concentration residual matrix to remove the information explained by the current latent variable;

[0232] Step 2.4

[0233] Generate regression model parameters:

[0234] a. Combine all latent variable weight vectors to form a projection matrix;

[0235] b. Generate a spectral regression coefficient vector based on the projection matrix and the blood glucose concentration weight coefficient;

[0236] c. Calculate the model intercept term to compensate for data centering bias;

[0237] (3) Data output stage

[0238] Define the blood glucose concentration prediction value as the linear combination of the new sample pretreatment spectrum vector and the regression coefficient vector plus the intercept term, and output the mapping relationship model.

[0239] The method for generating a corneal glucose concentration and health degree mapping relationship table comprises the following steps:

[0240] (1) Data acquisition stage:

[0241] In a controlled medical environment, simultaneously collect corneal reflectance spectrum data, corneal region temperature data, venous blood glucose concentration data and health status label data of the subjects;

[0242] Layer the subjects according to the preset health level:

[0243] Healthy group: fasting venous blood glucose 3.9-5.6 mmol / L;

[0244] Pre-diabetic group: fasting venous blood glucose 5.6-7.0 mmol / L;

[0245] Diabetes group: fasting venous blood glucose ≥7.0 mmol / L;

[0246] Generate a data set with a uniform timestamp for each group of subjects, with a timestamp accuracy of ≤1 millisecond;

[0247] (2) Data processing stage:

[0248] Extract the spectral feature vector;

[0249] Input the obtained spectral feature vector into the mapping relationship model obtained above, input the processed spectral feature vector into the initial model and perform iterative optimization to obtain the health risk prediction model

[0250] (3) Data output stage:

[0251] Construct a three-dimensional lookup table:

[0252] First dimension: corneal glucose concentration range 3.0-15.0 mmol / L;

[0253] Second dimension: metabolic fluctuation coefficient range 5%-50%;

[0254] Third dimension: age group [<30 years old, 30-60 years old, >60 years old];

[0255] Map the health risk prediction model output to the lookup table.

[0256] Reference the obtained corneal glucose concentration value to the preset control table representing the mapping relationship between corneal glucose concentration and health level, to obtain the health level of the target object.

[0257] When the environmental sensor fails or the interferometer fails, the LSTM model can be used to predict the tear film refractive index correction amount and the spectral compensation parameter according to the historical data. This model can still maintain the system function when the sensor or interferometer fails.

[0258] Detailed process of LSTM model activation

[0259] (1) Data acquisition stage:

[0260] Obtain the environmental temperature and humidity time series data, historical tear film thickness change rate (when available), sampling frequency: 10Hz (one sample every 100ms), time window length: n=10 (covering 1 second of historical data). All data are time-stamped synchronized through the PTP protocol, with a time deviation of less than 200μs. Construct a data set through the collected temperature and humidity time series data and historical tear film thickness change rate data.

[0261] (2) Data processing stage:

[0262] The structure of the neural network is constructed, including the input layer, hidden layer, and output layer. The Tensoflow framework is used to define the number of model neurons and activation functions.

[0263] The input dataset is passed through the network layers, layer by layer, with weight calculations and activation function processing, and finally generates a prediction result in the output layer.

[0264] The Huber loss function is used to improve the robustness of outliers, which is used to measure the error between the predicted value and the true value. By calculating the gradient of the loss function for each parameter, the error is propagated from the output layer to the input layer using the chain rule.

[0265] The optimizer uses the stochastic gradient descent (SGD) algorithm to update the model parameters based on the gradient calculated by backpropagation, accelerating convergence and improving training stability.

[0266] The model is trained through multiple iterations (epochs), and each iteration uses the training data for forward propagation, loss calculation, and parameter update. At the same time, the validation set is used to monitor the model performance to avoid overfitting.

[0267] After training is completed, the test set is used to evaluate the generalization ability of the model. By calculating accuracy, loss value, and other indicators, it is determined whether the model's performance on new data meets the normal fluctuation range of various indicators in the biological body or meets the system's requirement to remain continuous, smooth, and without sudden changes or jumps.

[0268] (3) Data output stage:

[0269] When the environmental sensor fails or the interferometer fails, the historical tear film change data and historical environmental temperature and humidity data are input into the LSTM model to predict the tear film thickness change rate at the next time, and the relevant historical environmental temperature and humidity data and the predicted tear film thickness change rate are substituted into the temperature and humidity compensation model and the refractive index correction formula to obtain the tear film refractive index correction amount and the spectral compensation parameter.

[0270] When the environmental sensor fails or the interferometer fails, a degradation strategy can be used to process the data: if the LSTM model predicts successfully, use the predicted value; if it fails, determine whether the historical data is available; if the historical data is available, use the most recent valid value; if the historical data is not available, return to the safe default value.

[0271] The details of each method step of the real-time non-invasive detection method of corneal glucose are described in detail above, and the application scenario of this embodiment is specifically described as follows:

[0272] Embodiment: Daily non-invasive blood glucose monitoring for diabetic patients

[0273] 1. Scene setting

[0274] Target subject: a 50-year-old male diabetic patient, who needs to monitor blood glucose level daily.

[0275] Environmental conditions: indoor home environment, temperature 28℃, humidity 60%.

[0276] Device configuration:

[0277] Optical sensing unit: near-infrared spectrometer (wavelength range 1050-1300 nm);

[0278] Biological measurement module: OCT (optical coherence tomography) tear film thickness measurement instrument;

[0279] Environmental monitoring unit: integrated temperature and humidity sensor (accuracy: ±0.1℃, ±1%).

[0280] 2. Implementation steps

[0281] Data acquisition:

[0282] The patient faces the optical probe, and the probe emits near-infrared light to the cornea at an incident angle of 30°.

[0283] Synchronous acquisition:

[0284] Corneal reflectance spectrum data (wavelength 1050 nm, 1300 nm);

[0285] Tear film thickness change data (sampling frequency 200 Hz);

[0286] Environmental temperature and humidity data (T=28℃, H=60%).

[0287] Model calculation:

[0288] Temperature and humidity compensation model:

[0289] Input: T=28℃, H=60%, T_ref=25℃, H_ref=40%;

[0290] Output: ΔS=α·(3)+β·(20)+γ·(3×20)=0.012;

[0291] Tear film refractive index correction model:

[0292] Input: tear film thickness change rate=0.02μm / s, T=28℃, H=60%;

[0293] Output: Δn=0.021×(0.02)0.0003×3+0.00015×20=0.00052;

[0294] n_t=1.336+0.00052=1.33652.

[0295] Joint correction:

[0296] Using joint correction formula:

[0297]

[0298] Corrected spectral data is used for subsequent glucose concentration calculation.

[0299] Data verification and replacement:

[0300] Retrieving reference spectrum under similar conditions (T=28℃, H=60%, tear film stable) from historical database.

[0301] Calculating deviation value between current spectrum and reference spectrum (mean deviation=0.08).

[0302] Fusing confidence (temperature and humidity compensation confidence=0.9, refractive index correction confidence=0.85) to get comprehensive confidence factor=0.875.

[0303] Adjusted deviation value=0.08 / 0.875≈0.091, which is less than threshold 0.1, and does not trigger replacement.

[0304] Glucose concentration and health level output:

[0305] Inputting corrected spectrum into PLS regression model, outputting glucose concentration=7.2mmol / L.

[0306] Querying health control table:

[0307] Glucose concentration 7.2mmol / L→diabetes group;

[0308] Outputting health level: “high risk, suggest medical review”.

[0309] Embodiment 2

[0310] The embodiment also provides a real-time non-invasive corneal glucose detection system, comprising:

[0311] A data acquisition module for acquiring corneal reflection spectrum data, tear film thickness change data of a target object, and temperature and humidity data of an environment where the target object is located in real time;

[0312] A data processing module for inputting the temperature and humidity data into a pre-constructed temperature and humidity compensation model to generate spectral compensation parameters and parameter confidence; and for inputting the temperature and humidity data and the tear film thickness change data into a pre-constructed refractive index correction model to generate tear film refractive index correction amount and correction amount confidence;

[0313] a joint correction processing module configured to perform joint correction on the reflection spectrum data by the joint correction formula using the spectral compensation parameter and the tear film refractive index correction amount; wherein, is real-time acquired reflection spectrum data, is a spectral compensation parameter, is a tear film refractive index correction amount, is a tear film reference refractive index;

[0314] a spectrum validity verification module configured to calculate reference reflection spectrum data based on a database constructed from historical reflection spectrum data, associated tear film thickness change data, and temperature and humidity data, and calculate a set of deviation values between the corrected reflection spectrum data and the reference reflection spectrum data, and extract statistical distribution characteristics of the set of deviation values as deviation values, further configured to weight and fuse the parameter confidence and the correction amount confidence to obtain a confidence factor, and further configured to adjust the deviation values based on the confidence factor, and if the deviation values are greater than a preset deviation threshold, replace the corrected reflection spectrum data with the reference reflection spectrum data;

[0315] a data analysis module configured to calculate a corneal glucose concentration value according to the processed reflection spectrum data, and then refer to a preset control table representing a mapping relationship between corneal glucose concentration and health level to obtain a health level of the target object.

[0316] Embodiment 3

[0317] The embodiment also provides a computing device, comprising:

[0318] a memory configured to store a program;

[0319] a processor configured to execute the computer executable instructions, and the computer executable instructions, when executed by the processor, implement the real-time non-invasive detection method for corneal glucose.

[0320] Embodiment 4

[0321] The embodiment also provides a computer readable storage medium storing a program, and the program, when executed by a processor, implements the real-time non-invasive detection method for corneal glucose.

[0322] The storage medium provided by the embodiment belongs to the same inventive concept as the real-time non-invasive detection method for corneal glucose provided by the above embodiments, and technical details not described in the embodiment can be referred to the above embodiments, and the embodiment has the same beneficial effects as the above embodiments.

[0323] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and necessary universal hardware, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application are essentially or say the contributions to the prior art

[0324] Part of it can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (Random Access Memory, RAM), FLASH, hard disk or optical disk, etc., including a number of instructions to make a computer device (may be a personal computer, server, or network equipment, etc.) execute the method of various embodiments of the present application.

[0325] The technical scope of the present application is not limited to the above description, and those skilled in the art can make various modifications and modifications to the above embodiments without departing from the technical idea of the present application, and these modifications and modifications should be within the scope of protection of the present application.

Claims

1. A method for real-time non-invasive detection of corneal glucose, characterized in that, include: Real-time acquisition of corneal reflectance spectrum data, tear film thickness change data, and temperature and humidity data of the target object's environment; The temperature and humidity data are input into a pre-built temperature and humidity compensation model to generate spectral compensation parameters and parameter confidence levels. The temperature and humidity data and tear film thickness change data are input into the pre-built refractive index correction model to generate the tear film refractive index correction amount and the confidence level of the correction amount. The spectral compensation parameters and tear film refractive index correction are used to jointly correct the reflectance spectral data using a joint correction formula; the joint correction formula is as follows: ,in, For real-time acquisition of reflectance spectral data, For spectral compensation parameters, This is the tear film refractive index correction amount. The reference refractive index is the tear film. A database built based on historical reflectance spectral data is used to calculate the baseline reflectance spectral data by linking tear film thickness variation data and temperature and humidity data. The set of deviation values ​​between the corrected reflectance spectral data and the baseline reflectance spectral data is calculated, and the statistical distribution characteristics of the deviation value set are extracted as the deviation value. The confidence factor is obtained by weighting and fusing the confidence scores of the parameters and the correction values. The deviation value is adjusted based on the confidence factor. If the deviation value is greater than the preset deviation threshold, the corrected reflectance spectral data is replaced with the reference reflectance spectral data. The corneal glucose concentration value is calculated based on the processed reflectance spectral data, and then a pre-defined reference table characterizing the mapping relationship between corneal glucose concentration and health status is used to obtain the health level of the target object.

2. The method for real-time non-invasive detection of corneal glucose according to claim 1, characterized in that, The construction of the temperature and humidity compensation model includes the following steps: The temperature and humidity of the environment where the target object is located, as well as the reflectance spectral data of the light source under different temperature and humidity and different concentrations of standard glucose solution, are obtained. The dataset is constructed after synchronizing the timestamps of the temperature and humidity and the reflectance spectral data. The dataset is then subjected to outlier handling, standardization, and data augmentation in sequence. The enhanced dataset is filtered using the random forest algorithm to obtain the processed dataset. An initial model is constructed based on reducing mean squared error and using a fully convolutional network. The processed dataset is then input into the initial model and iteratively optimized to obtain the temperature and humidity compensation model.

3. The method for real-time non-invasive detection of corneal glucose according to claim 1, characterized in that, The establishment of the refractive index correction model includes the following steps: Tear film thickness data of the sample subjects and the temperature and humidity of the environment in which the sample subjects were located were obtained; the sample subjects covered three groups of subjects: healthy people, diabetic patients and dry eye patients; After synchronizing all data with timestamps, a multimodal dataset is constructed by combining the target object category; The multimodal dataset is subjected to outlier handling, standardization, and data augmentation in sequence. Based on the random forest algorithm, the importance of features is selected, and the features of each sample object are sorted in descending order according to their weights. The top three features are retained to obtain the optimized multimodal dataset. A lightweight neural network is constructed based on the weighted MSE loss function and the stochastic gradient descent algorithm; The optimized multimodal dataset is input into a lightweight neural network for iterative training to obtain the refractive index correction model.

4. The method for real-time non-invasive detection of corneal glucose according to claim 1, characterized in that, The spectral compensation parameters, tear film refractive index correction, and reflectance spectral data are all timestamp-synchronized before joint correction.

5. The method for real-time non-invasive detection of corneal glucose according to claim 1, characterized in that, The corneal glucose concentration value was calculated based on the corrected reflectance spectrum data as follows: The corrected reflectance spectral data is input into the glucose concentration prediction model to generate corneal glucose concentration values. The method for constructing the glucose concentration prediction model includes the following steps: Acquire multiple sets of corneal reflectance spectrum datasets corresponding to known blood glucose concentrations in a controlled environment; The spectral data undergoes noise reduction filtering, baseline correction, and characteristic wavelength extraction. The processed spectral data is converted into digital signals, and spectral feature vectors are extracted. A mapping model between spectral feature vectors and blood glucose concentration is constructed by training a partial least squares regression algorithm. The spectral feature vectors are then input into the mapping model and iteratively optimized to obtain the glucose concentration prediction model.

6. The method for real-time non-invasive detection of corneal glucose according to claim 1, characterized in that, When the spectral compensation parameters and tear film refractive index correction are used to jointly correct the reflectance spectral data, the method also includes considering the device correction coefficient to modify the joint correction formula, as follows: Prepare standard solutions according to preset conditions; The theoretical and measured reflection intensities of the standard solution at the reference wavelength are obtained, and all data are timestamped. The equipment correction coefficient is obtained by calculating the ratio of the two. ; Equipment correction factor The joint correction formula is modified to obtain the modified joint correction formula: .

7. The method for real-time non-invasive detection of corneal glucose according to claim 1, characterized in that, When acquiring corneal reflectance spectrum data, tear film thickness change data, and temperature and humidity data of the target object's environment in real time, if the tear film thickness change data and temperature and humidity data of the target object's environment are not acquired, the historical tear film change data and environmental temperature and humidity data are input into the LSTM model to predict the tear film thickness change data and temperature and humidity data at the current moment.

8. A real-time non-invasive corneal glucose detection system, characterized in that, The system is executed by the real-time non-invasive corneal glucose detection method according to any one of claims 1-7, and the system comprises: The data acquisition module is used to acquire corneal reflectance spectrum data, tear film thickness change data, and temperature and humidity data of the environment in which the target object is located in real time. The data processing module is used to input the temperature and humidity data into a pre-built temperature and humidity compensation model to generate spectral compensation parameters and parameter confidence levels; it is also used to input the temperature and humidity data and tear film thickness change data into a pre-built refractive index correction model to generate tear film refractive index correction amount and correction amount confidence levels. The joint correction processing module is used to jointly correct the reflectance spectral data by combining the spectral compensation parameters and the tear film refractive index correction amount using a joint correction formula; the joint correction formula is: ,in, For real-time acquisition of reflectance spectral data, For spectral compensation parameters, This is the tear film refractive index correction amount. The reference refractive index is the tear film. The spectral validity verification module is used to construct a database based on historical reflectance spectral data. It calculates baseline reflectance spectral data by linking tear film thickness variation data and temperature and humidity data, and calculates the set of deviation values ​​between the corrected reflectance spectral data and the baseline reflectance spectral data. It also extracts the statistical distribution characteristics of the deviation value set as the deviation value. Furthermore, it is used to weight and fuse parameter confidence and correction confidence to obtain a confidence factor. Finally, it is used to adjust the deviation value based on the confidence factor. If the deviation value is greater than a preset deviation threshold, the corrected reflectance spectral data is replaced with the baseline reflectance spectral data. The data analysis module calculates the corneal glucose concentration value based on the processed reflectance spectral data, and then refers to a preset reference table that characterizes the mapping relationship between corneal glucose concentration and health status to obtain the health level of the target object.

9. An electronic device, characterized in that, include: Memory, used to store programs; A processor for loading the program to perform the steps of the real-time non-invasive corneal glucose detection method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the steps of the real-time non-invasive corneal glucose detection method as described in any one of claims 1-7.

Images