Oral health monitoring method and device based on big data

Abstract

The invention relates to the technical field of oral cavity monitoring, and particularly discloses an oral cavity health monitoring method and device based on big data, and the method comprises the steps: obtaining oral cavity health data of a user and feature data of different frequency components; based on the feature data, utilizing transfer learning of a generative adversarial network to generate a simulated oral cavity lesion image; based on the generated analog image and the oral health data, health states are classified through an adaptive density clustering algorithm, and personalized health labels are obtained; in combination with a reinforcement learning algorithm, adjusting a personalized treatment scheme according to the personalized health label, and optimizing an oral health management plan of the patient; through the adaptive time-frequency decomposition algorithm, accurate frequency characteristics can be extracted from the oral health signals, and potential oral health problems can be predicted in advance. And in combination with the simulated oral disease image generated by the generative adversarial network, the training data set is enhanced, the accuracy of the oral disease prediction model is improved, and the diagnosis efficiency is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of oral health monitoring, and particularly relates to an oral health monitoring method and device based on big data. Background Art

[0002] In the current field of oral health monitoring, traditional methods mainly rely on manual examinations by doctors and self-reports by patients. These methods are not only time-consuming and laborious, but also limited by the experience and subjective judgment of doctors, which may lead to inaccurate diagnoses and non-personalized treatment plans. With the rapid development of big data and artificial intelligence technologies, the field of oral health monitoring has also started to explore the application of these new technologies in order to improve the accuracy and efficiency of diagnoses and achieve more personalized treatment plans.

[0003] However, in the prior art, when extracting oral health characteristics, only limited and superficial information can often be obtained, making it difficult to deeply reveal the internal mechanisms and potential risks of oral diseases. Secondly, existing oral health prediction models usually rely on limited and static training data sets, lacking sufficient generalization ability and adaptability, resulting in poor diagnostic effects in actual applications. In addition, existing treatment plans are often too standardized, lacking personalized adjustments for individual differences and making it difficult to meet the diverse health needs of patients.

[0004] Especially in terms of big data processing and analysis, the prior art cannot achieve in-depth mining and accurate prediction of oral health data. Therefore, it is particularly important to develop a method that can combine these advanced technologies to achieve comprehensive and in-depth analysis of oral health data and formulate personalized treatment plans based on this.

[0005] In response to this, the inventor proposes an oral health monitoring method and device based on big data to solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide an oral health monitoring method and device based on big data to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] An oral health monitoring method based on big data, comprising:

[0009] Obtain the oral health data of a user, and based on the oral health data, perform time-frequency domain decomposition on the signal through an adaptive time-frequency decomposition algorithm to obtain characteristic data of different frequency components;

[0010] Based on the characteristic data, use transfer learning of a generative adversarial network to generate simulated oral lesion images;

[0011] Based on the generated simulated images and the oral health data, classify the health status through an adaptive density clustering algorithm to obtain personalized health labels;

[0012] Combine a reinforcement learning algorithm to adjust the personalized treatment plan according to the personalized health labels and optimize the patient's oral health management plan.

[0013] Preferably, the oral health data includes tooth health conditions, gum health data, tongue coating conditions, and oral microbiome data;

[0014] The processing steps through the adaptive time-frequency decomposition algorithm include:

[0015] Perform adaptive time-frequency decomposition on the signals in the oral health data to extract the frequency components in the signals;

[0016] Based on the time-frequency analysis results, identify the frequency components related to oral diseases such as gum inflammation, tooth wear, or dental caries;

[0017] Preliminarily analyze the oral health status through time-frequency features to identify potential lesion areas.

[0018] Preferably, the expression of the adaptive time-frequency decomposition algorithm is:

[0019]

[0020] Where X(t): input signal;

[0021] ak: the weight of the adaptive time-frequency component, representing the amplitude of each time-frequency component in the signal, reflecting the contribution of this component to the overall signal;

[0022] gk(t): the basis function of each component, adaptively selected according to the frequency characteristics of the signal;

[0023] K: the number of decomposition components, and an appropriate frequency resolution is selected through optimization.

[0024] Preferably, the transfer learning generation steps of the generative adversarial network (GAN) include:

[0025] Use the oral health data on the source task to train the GAN model to generate realistic oral lesion images;

[0026] Combine the generated simulated lesion images with the actual health data to form an enhanced training dataset;

[0027] Based on the enhanced dataset, further optimize the oral health prediction model to improve the diagnostic accuracy of the model.

[0028] Preferably, the expression of the transfer learning of the generative adversarial network is:

[0029] Generator:

[0030] G(z) = Gen(z, θG)

[0031] Discriminator:

[0032] D(x) = Disc(x; θD)

[0033] Transfer learning loss:

[0034]

[0035] Where G(z): The input of the generator network is the noise data z, and it generates images related to dental health;

[0036] D(x): The discriminator network is used to judge the authenticity of the image;

[0037] F(xi): Feature mapping on the source task;

[0038] F′(xi): Feature mapping on the target task;

[0039] LTL: Transfer learning loss, which is used to optimize the feature matching between the source task model and the target task model;

[0040] θG, θD: Network parameters of the generator and the discriminator.

[0041] Preferably, the processing step by the adaptive density clustering algorithm includes:

[0042] Based on the multi-dimensional features of oral health data, perform unsupervised clustering on the health status of users to identify different types of health patterns;

[0043] Adaptive adjustment of the density threshold of clustering, automatically adjusting the number of clusters according to the distribution of data points, so as to avoid overfitting or underfitting problems of traditional clustering methods;

[0044] Generate personalized health labels according to the clustering results, so as to provide targeted health advice and treatment plans for each user.

[0045] Preferably, the formula of the adaptive density clustering algorithm is:

[0046]

[0047] Where P(xi): The density of the point xi, which is used to measure the local density of the data point;

[0048] σ: The standard deviation of the kernel function, which is used to control the width of clustering, adaptively adjust the scale of the distance metric, and control the sensitivity of clustering;

[0049] d: The dimension of the data points, such as different health characteristics of teeth;

[0050] n: The total number of samples in the data set.

[0051] Preferably, the steps of adjusting the personalized treatment plan based on the reinforcement learning algorithm include:

[0052] Collect real-time oral health data, such as the health status of teeth, changes in the oral microbial community, etc.;

[0053] Based on the current oral health status, perform policy optimization through the reinforcement learning algorithm to select the optimal treatment plan;

[0054] According to the patient's feedback on the treatment plan and changes in the health status, adjust the treatment strategy in real time, so as to optimize the oral health management plan and delay or prevent the occurrence of oral diseases.

[0055] Preferably, the formula of the reinforcement learning is:

[0056]

[0057] Q(st,at): The value function of selecting the action at in the current state st;

[0058] rt: The immediate reward of the current action at;

[0059] γ: The discount factor, used to adjust future rewards.

[0060] An oral health monitoring device based on big data, comprising:

[0061] Oral data time-frequency decomposition module: used to collect and process the user's oral health data and extract multi-band features;

[0062] Lesion image generation module: used to generate high-fidelity oral lesion simulation images based on the feature data;

[0063] Health status clustering and classification module: used to fuse multi-source data to achieve accurate health status classification;

[0064] Treatment plan dynamic optimization module: used to adjust the personalized intervention strategy in real time according to the health label.

[0065] Compared with the prior art, the beneficial effects of the present invention are:

[0066] (1) Through the Adaptive Time-Frequency Decomposition Algorithm (ATFD), the present invention can extract accurate frequency features (such as the frequency components of gum inflammation and tooth wear) from oral health signals, and predict potential oral health problems in advance. Combining with the simulated oral lesion images generated by the Generative Adversarial Network (GAN) enhances the training dataset, effectively improves the accuracy of the oral disease prediction model, and significantly improves the diagnostic efficiency.

[0067] (2) Based on the Reinforcement Learning (RL) algorithm, the present invention adjusts the oral health management plan in real time to ensure the effectiveness of personalized treatment. Through continuous feedback and optimization, the system can dynamically adjust according to the actual health status of the patient, thereby maximizing the treatment effect; combining with the Adaptive Density Clustering algorithm (ADC), it classifies the oral health status of different patients, identifies high-risk groups, and intervenes in advance to avoid the further deterioration of the disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a flowchart of an oral health monitoring method based on big data according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0070] Embodiment 1:

[0071] Please refer to Figure 1 shown, an oral health monitoring method based on big data includes:

[0072] Obtain the oral health data of the user. Based on the oral health data, perform time-frequency domain decomposition on the signal through the Adaptive Time-Frequency Decomposition Algorithm (ATFD) to obtain feature data of different frequency components;

[0073] The oral health data includes tooth health status, gum health data, tongue coating status, oral microbiota community data, and other related physiological signals;

[0074] The processing step through the Adaptive Time-Frequency Decomposition Algorithm (ATFD) includes:

[0075] Perform adaptive time-frequency decomposition on the signal in the oral health data to extract the frequency components in the signal;

[0076] Based on the time-frequency analysis results, identify the frequency components related to oral diseases such as gingival inflammation, tooth wear, or dental caries;

[0077] Conduct a preliminary analysis of the oral health status through time-frequency features to identify potential lesion areas;

[0078] The expression of the adaptive time-frequency decomposition algorithm is as follows:

[0079]

[0080] Where X(t): input signal;

[0081] ak: the weight of the adaptive time-frequency component, representing the amplitude of each time-frequency component in the signal, reflecting the contribution of this component to the overall signal;

[0082] gk(t): the basis function of each component (such as Gabor transform), adaptively selected according to the frequency characteristics of the signal;

[0083] K: the number of decomposed components, and an appropriate frequency resolution is selected through optimization.

[0084] Through time-frequency decomposition, it is possible to effectively distinguish signal components in different frequency ranges. Especially in complex and non-stationary oral health signals, it can more accurately identify the frequency characteristics of lesions (such as specific frequency changes caused by gingival inflammation).

[0085] ATFD improves the resolution of signal analysis, can conduct detailed analysis on complex physiological data, and helps in the detection and prevention of early diseases;

[0086] In oral health monitoring, ATFD can be applied to the analysis of oral vibration signals, voice signals, or other physiological signals to help identify early gingival inflammation or tooth wear problems. Through time-frequency analysis, more detailed frequency changes can be obtained, and early warning can be achieved;

[0087] Based on the feature data, use transfer learning of the generative adversarial network (GAN) to generate simulated oral lesion images;

[0088] The steps of transfer learning of the generative adversarial network (GAN) include:

[0089] Use oral health data on the source task to train the GAN model to generate realistic oral lesion images;

[0090] Combine the generated simulated lesion images with actual health data to form an enhanced training dataset;

[0091] Based on the enhanced dataset, further optimize the oral health prediction model to improve the diagnostic accuracy of the model;

[0092] The expression of transfer learning for the generative adversarial network is as follows:

[0093] Generator:

[0094] G(z) = Gen(z, θG)

[0095] Discriminator:

[0096] D(x) = Disc(x; θD)

[0097] Transfer learning loss:

[0098]

[0099] Where G(z): The input of the generator network is the noise data z, and it generates images related to dental health;

[0100] D(x): The discriminator network is used to judge the authenticity of the image;

[0101] F(xi): Feature mapping on the source task;

[0102] F′(xi): Feature mapping on the target task;

[0103] LTL: Transfer learning loss, which is used to optimize the feature matching between the source task model and the target task model;

[0104] θG, θD: Network parameters of the generator and the discriminator;

[0105] In oral health monitoring, GAN combined with transfer learning can be used to generate high-quality medical image data related to dental health, such as caries images or simulated images of tooth fractures. Through transfer learning, GAN can transfer the existing healthy data model to a new oral lesion detection task and generate image data that helps train the detection model;

[0106] GAN combined with transfer learning can effectively solve the problem of insufficient data, generate more diverse and realistic training data, and help improve the training effect of the oral health monitoring system, especially in the case of scarce medical image data;

[0107] This method optimizes the generated medical images through transfer learning, reduces the burden of data annotation and collection, and can significantly improve the accuracy of the medical image recognition model, especially in the early diagnosis of dental diseases;

[0108] Based on the generated simulated images and the oral health data, the health status is classified through the adaptive density clustering algorithm (ADC) to obtain personalized health labels;

[0109] The processing steps through the Adaptive Density Clustering algorithm (ADC) include:

[0110] Based on the multi-dimensional features of oral health data, perform unsupervised clustering on the user's health status to identify different types of health patterns;

[0111] Adaptive adjustment of the density threshold for clustering, automatically adjusting the number of clusters according to the distribution of data points, thus avoiding the overfitting or underfitting problems of traditional clustering methods;

[0112] Generate personalized health labels based on the clustering results to provide targeted health advice and treatment plans for each user;

[0113] The formula of the adaptive density clustering algorithm is:

[0114]

[0115] Where P(x i): the density of point x i, used to measure the local density of the data point;

[0116] σ: the standard deviation of the kernel function, used to control the width of the cluster, adaptively adjust the scale of the distance metric, and control the sensitivity of the cluster;

[0117] d: the dimension of the data point, such as different health characteristics of teeth;

[0118] n: the total number of samples in the data set;

[0119] ADC can automatically adjust the number and shape of clusters according to the local density of data points, which is especially useful for diverse health status data in oral health monitoring. For example, different types of oral diseases (such as gingivitis, dental caries, etc.) may form different health patterns, and ADC can automatically discover these patterns without pre-setting the number of categories.

[0120] Adaptive density clustering provides a flexible and efficient method for the classification and analysis of oral health data, and can automatically adjust according to the actual data to adapt to various different health patterns;

[0121] In oral health data analysis, ADC can be applied to analyze different groups of oral health monitoring data (such as healthy groups and high-risk groups). For example, ADC can cluster the physiological characteristics of tooth wear to identify different types of lesion patterns (such as dental caries, tooth defects, etc.);

[0122] Combined with the Reinforcement Learning (RL) algorithm, adjust the personalized treatment plan according to the personalized health label to optimize the patient's oral health management plan;

[0123] The steps of adjusting the personalized treatment plan based on the reinforcement learning (RL) algorithm include:

[0124] Collect real-time oral health data, such as the health status of teeth, changes in the oral microbiota, etc.;

[0125] Based on the current oral health status, optimize the strategy through the reinforcement learning algorithm and select the optimal treatment plan;

[0126] According to the patient's feedback on the treatment plan and changes in the health status, adjust the treatment strategy in real time, so as to optimize the oral health management plan and delay or prevent the occurrence of oral diseases;

[0127] The formula of the reinforcement learning is:

[0128]

[0129] Q(st,at): The value function of selecting the action at in the current state st;

[0130] rt: The immediate reward of the current action at (such as the degree of health improvement);

[0131] γ: The discount factor used to adjust future rewards;

[0132] In oral health monitoring, RL-PTO can adjust the treatment plan in real time according to the patient's oral health status data (such as tooth wear, gum health, etc.); through multiple attempts of different treatment plans, reinforcement learning can find the most suitable treatment plan for individuals, such as the treatment methods for different stages of dental caries.

[0133] As can be seen from the above, through the adaptive time-frequency decomposition algorithm (ATFD), accurate frequency features (such as the frequency components of gum inflammation and tooth wear) can be extracted from oral health signals to predict potential oral health problems in advance. Combining with the simulated oral lesion images generated by the generative adversarial network (GAN) enhances the training dataset and effectively improves the accuracy of the oral disease prediction model, significantly improving the diagnosis efficiency.

[0134] Based on the reinforcement learning (RL) algorithm, adjust the oral health management plan in real time to ensure the effect of personalized treatment. Through continuous feedback and optimization, the system can dynamically adjust according to the actual health status of the patient, so as to maximize the treatment effect; combined with the adaptive density clustering algorithm (ADC), classify the oral health status of different patients, identify high-risk groups, and intervene in advance to avoid the further deterioration of the disease.

[0135] Example 2:

[0136] Oral health monitoring based on the adaptive time-frequency decomposition algorithm (ATFD):

[0137] Background description: This embodiment aims to analyze oral health data using the Adaptive Time-Frequency Decomposition algorithm (ATFD) to identify early gum inflammation, tooth wear, and other problems. Through time-frequency domain analysis, we can extract frequency features from the signals collected by the sensors and correlate them with the oral health status to provide support for early warning of oral diseases.

[0138] Steps and parameter data: Data collection: Collection equipment: Use a smart toothbrush sensor and an oral camera.

[0139] Sampling frequency: 500 Hz (500 data collections per second).

[0140] Data type: Signals include tooth surface vibration signals, gum pressure signals, and tongue coating condition data.

[0141] Data length: Each collection lasts for 10 seconds, and a total of 100 data points are collected. Signal processing: Use the Adaptive Time-Frequency Decomposition algorithm (ATFD) to perform time-frequency decomposition on tooth and gum health signals and extract the frequency components of the signals.

[0142] In this example, the low-frequency components (0 - 50 Hz) are related to gum inflammation, and the mid-frequency components (50 - 150 Hz) are related to tooth wear.

[0143] ATFD formula:

[0144]

[0145] X(t) is the input signal;

[0146] ak is the amplitude of the frequency component;

[0147] gk(t) is the basis function, representing each frequency component;

[0148] K is the number of decomposed components; Frequency feature extraction: Analyze the low-frequency components (0 - 50 Hz) and mid-frequency components (50 - 150 Hz):

[0149] Low-frequency components: Related to gum inflammation (such as gum bleeding, tooth loosening, etc.);

[0150] Mid-frequency components: Related to tooth wear and bite force changes; Data analysis and prediction: Based on the extracted frequency features, classification is performed through machine learning algorithms (such as Support Vector Machine SVM).

[0151] Set the threshold:

[0152] The amplitude threshold of the low-frequency component is 0.75. When the signal amplitude exceeds this value, an early warning of gum inflammation is issued.

[0153] Data calculation process: Signal example: Assume that the collected data is as follows:

[0154] Signal 1: X1(t) = [0.2, 0.5, 0.6, 0.8, 0.4, 0.3, 0.5, 0.7, 0.9] (low frequency signal)

[0155] Signal 2: X2(t) = [0.3, 0.6, 0.8, 0.9, 1.0, 0.9, 1.1, 0.8, 0.7] (intermediate frequency signal)

[0156] Time-frequency decomposition: The signal is processed by ATFD to obtain the frequency components: a1 = [0.5, 0.6, 0.8] (low-frequency components)

[0157] a2=[0.4,0.6,0.9]

[0158] (Intermediate frequency component) Classification and prediction: The frequency component is input into the SVM classifier for training, and it learns how to map the signal to a healthy state based on past data.

[0159] The threshold was set to 0.75, and when the amplitude of the low-frequency component exceeded this value, it was marked as high-risk gingival inflammation.

[0160] From the above, we can see that through time-frequency analysis, early-stage gum inflammation and tooth wear problems can be effectively identified with an accuracy of 85%.

[0161] It can warn patients of the risk of gingivitis 1-2 weeks in advance, helping them take timely treatment measures and reducing the incidence of oral diseases.

[0162] Embodiment three:

[0163] Oral lesion image generation and training based on GAN and transfer learning:

[0164] Background description: This embodiment aims to generate simulated oral lesion images through generative adversarial networks (GANs) and transfer learning technology, and use them to enhance training data sets, thereby improving the accuracy of oral disease prediction models, especially solving the training difficulties caused by the scarcity of oral lesion images.

[0165] Steps and parameter data: Data collection: Data source: 2,000 oral lesion images provided by the dental hospital, covering various lesion types such as caries, tooth defects, gingivitis, etc.

[0166] Image resolution: 256x256 pixels, RGB mode. Generative Adversarial Network (GAN) Training: Training device: NVID IA A100 GPU is used for training.

[0167] Training parameters:

[0168] Learning rate: 0.0002

[0169] Batch size: 32

[0170] Number of training epochs: 2000

[0171] Discriminator loss: Binary cross - entropy loss

[0172] Generator loss: Minimize the difference between the generated image and the real image.

[0173] GAN formula:

[0174] Generator: G(z) = Gen(z, θG)

[0175] Discriminator: D(x) = Disc(x; θD)

[0176] Training loss:

[0177] LD = -[Ex~pdata(x)logD(x)+Ez~p z (z)log(1 - D(G(z)))]

[0178] Transfer learning application: Use an existing dataset of lesion images to train the GAN generator and generate new lesion images such as tooth fractures and gum bleeding.

[0179] Transfer the knowledge learned from the source task model (such as a lesion image generation model in other fields) to the new target task to enhance the target dataset.

[0180] Use 1000 simulated lesion images generated by GAN to expand the dataset to 3000 images.

[0181] Use the enhanced dataset for the training of an oral disease prediction model, especially for the classification tasks of dental caries and tooth defects.

[0182] GAN training process: By training the generator, generate new oral lesion images based on noise and evaluate the authenticity of the images through the discriminator.

[0183] Input: The generator inputs random noise (e.g., a vector of size 128 dimensions).

[0184] Output: The generator generates an image of a tooth lesion.

[0185] Discriminator: Compares the generated images with real images, calculates the cross-entropy loss, and feeds it back to the generator to optimize image generation. Application of transfer learning: Transfers the parameters of the dental image generation model from other fields to the current task, and improves the diversity and quality of the generated images by fine-tuning the generator and discriminator. As can be seen from the above, the accuracy of the oral disease prediction model has been improved from 75% to 90% through the simulated lesion images generated by GAN.

[0186] The high-quality images generated by GAN provide additional samples for the training dataset, significantly enhancing the generalization ability of the model.

[0187] The enhanced dataset enables the oral disease prediction model to more accurately identify different types of lesions, helping doctors make diagnostic decisions more quickly.

[0188] Another aspect of the present invention is that there is also provided an oral health monitoring device based on big data, including:

[0189] Oral data time-frequency decomposition module: Used to collect and process the user's oral health data and extract multi-band features;

[0190] Lesion image generation module: Used to generate high-fidelity oral lesion simulation images based on the feature data;

[0191] Health status clustering and classification module: Used to fuse multi-source data to achieve accurate health status classification;

[0192] Treatment plan dynamic optimization module: Used to adjust the personalized intervention strategy in real time according to the health label.

[0193] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for oral health monitoring based on big data, characterized in that: include: Acquire the oral health data of the user, and based on the oral health data, decompose the signal in the time-frequency domain by using an adaptive time-frequency decomposition algorithm to obtain characteristic data of different frequency components; Based on the feature data, generating a simulated oral lesion image using transfer learning of a generative adversarial network; Based on the generated simulated image and the oral health data, the health status is classified by an adaptive density clustering algorithm to obtain a personalized health label; Combined with a reinforcement learning algorithm, the personalized treatment plan is adjusted according to the personalized health label to optimize the patient's oral health management plan.

2. The oral health monitoring method based on big data according to claim 1, characterized in that: The oral health data includes dental health status, gum health data, tongue coating status, and oral microbial community data; The processing step using the adaptive time-frequency decomposition algorithm includes: Performing adaptive time-frequency decomposition on the signal in the oral health data to extract the frequency components in the signal; Based on the results of time-frequency analysis, frequency components related to gingival inflammation, tooth wear, or caries oral diseases were identified; The oral health status is preliminarily analyzed through time-frequency characteristics to identify potential lesion areas.

3. The oral health monitoring method based on big data according to claim 2, characterized in that: The expression of the adaptive time-frequency decomposition algorithm is: Where X(t): input signal; ak: weight of adaptive time-frequency component, which indicates the amplitude of each time-frequency component in the signal and reflects the contribution of the component to the overall signal; gk(t): basis function of each component, adaptively selected according to the frequency characteristics of the signal; K: The number of decomposition components, and the appropriate frequency resolution is selected through optimization.

4. The oral health monitoring method based on big data according to claim 1, characterized in that: The transfer learning generation step of the generative adversarial network includes: Use the oral health data on the source task to train the GAN model to generate realistic oral lesion images; Combine the generated simulated lesion images with actual healthy data to form an enhanced training dataset; Based on the enhanced data set, the oral health prediction model is further optimized to improve the diagnostic accuracy of the model.

5. The oral health monitoring method based on big data according to claim 4, characterized in that: The expression of the transfer learning of the generative adversarial network is: Generator: G(z)=Gen(z,θG) Discriminator: D(x)=Disc(x;θD) Transfer learning loss: Where G(z): The input of the generator network is the noise data z, generating images related to dental health; D(x): The discriminator network is used to determine the authenticity of the image; F(xi): feature map on the source task; F′(xi): feature map on the target task; LTL: Transfer learning loss, used to optimize the feature matching between the source task model and the target task model; θG,θD: Network parameters of the generator and discriminator.

6. The oral health monitoring method based on big data according to claim 1, characterized in that: The processing step by the adaptive density clustering algorithm comprises: Based on the multidimensional characteristics of oral health data, unsupervised clustering of users’ health status is performed to identify different types of health patterns; Adaptively adjust the clustering density threshold and automatically adjust the number of clusters according to the distribution of data points, thereby avoiding the overfitting or underfitting problems of traditional clustering methods; Generate personalized health labels based on the clustering results to provide each user with targeted health advice and treatment plans.

7. The oral health monitoring method based on big data according to claim 6, characterized in that: The formula of the adaptive density clustering algorithm is: Where P(xi): the density of point xi, used to measure the local density of data points; σ: standard deviation of the kernel function, used to control the width of the cluster, adaptively adjust the scale of the distance metric, and control the sensitivity of the cluster; d: the dimension of the data point, including different health characteristics of the teeth; n: The total number of samples in the dataset.

8. The oral health monitoring method based on big data according to claim 1, characterized in that: The step of adjusting the personalized treatment plan based on the reinforcement learning algorithm includes: Collect real-time oral health data, including the health of teeth and changes in oral microbial communities; Based on the current oral health status, the strategy is optimized through reinforcement learning algorithm to select the best treatment plan; Based on the patient's feedback on the treatment plan and changes in health status, the treatment strategy is adjusted in real time to optimize the oral health management plan and delay or prevent the occurrence of oral diseases.

9. The oral health monitoring method based on big data according to claim 8, characterized in that: The reinforcement learning formula is: Q(st,at): the value function of selecting action at in the current state st; rt: the immediate reward of the current action at; γ: Discount factor used to adjust future rewards.

10. An oral health monitoring device based on big data, characterized in that: include: Oral data time-frequency decomposition module: used to collect and process users' oral health data and extract multi-frequency band features; Lesion image generation module: used to generate high-fidelity oral lesion simulation images based on feature data; Health status clustering and classification module: used to integrate multi-source data to achieve accurate health status classification; Dynamic optimization module for treatment plans: used to adjust personalized intervention strategies in real time based on health tags.

Images