Periodontal tissue health state monitoring method based on oral cavity detection indexes
Through the extreme learning machine algorithm dynamically adjusting the neuron response mode and feature reconstruction, the local optimal solution problems caused by data noise and sample in periodontal tissue health status monitoring are solved, and more efficient classification and stability are achieved.
Patent Information
- Application Number
- CN202510549247.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-28
- Publication Date
- 2025-08-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When handling periodontal tissue health status monitoring, existing machine learning algorithms are susceptible to data noise and sample inhomogeneity, resulting in falling into local optimal solutions and unsatisfactory classification results.
The extreme learning machine algorithm is used to initialize weights and bias parameters, calculate the entropy rate, dynamically adjust the neuron response mode, combine feature reconstruction and nonlinear activation functions, optimize the mapping space, avoid local optimal solutions and reduce feature redundancy.
It improves the classification accuracy of periodontal tissue health status monitoring and the stability of the model, enhances the adaptability and robustness of complex oral detection data, reduces feature redundancy, and improves data distinction.
Smart Images
Figure CN120511046A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of oral cavity detection, and in particular relates to a method for monitoring the health status of periodontal tissue based on oral cavity detection indicators. Background Art
[0002] With the improvement of people's living standards, oral health issues have gradually become an important concern in the field of public health, especially the incidence of diseases such as periodontitis has increased year by year. Periodontal health directly affects overall health, and its development process is relatively slow, and there are often no obvious symptoms in the early stages, so early detection and intervention are crucial. At present, the diagnosis of periodontal disease mainly relies on oral examination and clinical evaluation. However, these traditional methods usually rely on manual judgment and are limited by the doctor's experience and time, making it difficult to achieve efficient and accurate early diagnosis. With the advancement of technology, oral health monitoring is gradually moving towards automation and data-driven directions. Data acquisition systems based on various oral detection indicators (such as periodontal pocket depth, gingival bleeding index, etc.) are becoming increasingly perfect.
[0003] Existing technologies still face challenges in data collection and processing, feature extraction, and analysis. For example, oral examination data typically has high dimensionality and complex features, which makes traditional analysis methods prone to feature redundancy and data imbalance during processing, affecting model training efficiency and classification accuracy. Furthermore, when data is noisy or the sample size is uneven, existing machine learning algorithms often fall into local optimal solutions, resulting in suboptimal classification results. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for monitoring the health status of periodontal tissues based on oral detection indicators, aiming to solve the problem that existing machine learning algorithms often fall into local optimal solutions when data noise is large or samples are uneven, resulting in unsatisfactory classification results.
[0005] The present invention is implemented as follows: a method for monitoring the health status of periodontal tissue based on oral detection indicators, the method comprising:
[0006] Initialize the extreme learning machine's weights, bias parameters, and information density adjustment items, and construct the extreme learning machine's weight expression;
[0007] Calculate the entropy rate of each neuron in the extreme learning machine and use the entropy rate to measure the rate of change of oral detection data;
[0008] During each training iteration, the weights and biases are updated and adjusted through a dynamic random jump mechanism;
[0009] The hidden layer of the extreme learning machine contains a feature reconstruction module, which reconstructs features by analyzing the correlation between each feature in the training oral examination data and optimizes the mapping space;
[0010] In the hidden layer output stage of the extreme learning machine, the connection weights between neurons are adjusted through back propagation, and the hidden layer output of the extreme learning machine is calculated;
[0011] In the output layer of the extreme learning machine, a nonlinear activation function is used and iterations are repeated until the stopping condition is met. The health status of periodontal tissue is monitored through the trained extreme learning machine.
[0012] Preferably, the weight of the extreme learning machine is expressed as:
[0013]
[0014] Where W u is the weight of the extreme learning machine; ~ means obeying a specific distribution; n u is the feature number of oral detection data input to the extreme learning machine; α ur is the weight scaling factor, which is used to control the amplitude of weight initialization; Indicates that the mean is 0 and the variance is σ ua 2 Normal distribution; represents normal distribution; σ ua 2 is the initialization variance of the extreme learning machine weight parameters; u Represents the information density adjustment factor based on the characteristics of the input oral detection data.
[0015] Preferably, the information density adjustment factor is expressed as:
[0016]
[0017] Where, X c,i represents the i-th eigenvalue of the oral cavity detection data input by the extreme learning machine; X c is the oral cavity detection data input to the extreme learning machine; || || is the L2 norm; τ u An exponential factor to control the impact of information density; is the initialization variance of the information density adjustment factor; i is a positive integer.
[0018] Preferably, the entropy rate of a neuron is expressed as:
[0019]
[0020] Where H uea represents the entropy rate of the neurons of the extreme learning machine; p irepresents the output response probability of the i-th neuron of the extreme learning machine; log() represents the natural logarithm function; m ur is the number of neurons in the extreme learning machine; λ3 is the mutual information regularization coefficient; I() represents the mutual information calculation function; H u is the hidden layer output of the extreme learning machine.
[0021] Preferably, in the step of adjusting the weight and bias update by the dynamic random jump mechanism, the adjustment amount of the extreme learning machine weight is expressed as:
[0022]
[0023] In the formula, ΔW u Represents the adjustment amount of the extreme learning machine weight for each jump; β1 is the jump scale factor; ∈ u is the jump factor strength; represents jump initialization variance parameter;
[0024] The update process of the extreme learning machine weight is expressed as:
[0025]
[0026] Where, Represents the updated extreme learning machine weights.
[0027] Preferably, in the step of adjusting the weight and bias updates by a dynamic random jump mechanism, an adaptive jump strength is adopted to dynamically adjust the jump factor strength according to the current loss function gradient. The calculation method of the jump factor strength is expressed as:
[0028]
[0029] Where η u represents the basic jump strength; γ1 is the adaptive adjustment factor; Represents the gradient of the extreme learning machine's loss function with respect to the weights.
[0030] Preferably, the hidden layer of the extreme learning machine includes a feature reconstruction module, which reconstructs the features by analyzing the correlation between the features in the training oral detection data and optimizing the mapping space. The feature after feature reconstruction is represented as:
[0031]
[0032] Where, Represents the features reconstructed by the feature reconstruction module; X c represents the input oral cavity detection data of the feature reconstruction module; M u Represents the feature mapping matrix, which determines how the input features are mapped to the new feature space.
[0033] Preferably, the feature mapping matrix is optimized according to the reconstruction error gradient, and the updating method of the feature mapping matrix is expressed as:
[0034]
[0035] Where, Represents the updated feature mapping matrix; λ2 is the learning rate of mapping optimization; Represents the reconstruction error E u About the feature mapping matrix M u gradient; represents the noise error term.
[0036] Preferably, in the step of adjusting the connection weights between neurons by back propagation and calculating the hidden layer output of the extreme learning machine in the hidden layer output stage of the extreme learning machine, the hidden layer output of the extreme learning machine is expressed as:
[0037]
[0038] Where H u is the hidden layer output of the extreme learning machine; Sig() is the Sigmoid activation function; represents the updated extreme learning machine weight; Represents the features reconstructed by the feature reconstruction module.
[0039] Preferably, in the output layer of the extreme learning machine, in the process of using a nonlinear activation function, the nonlinear mapping result of the output layer of the extreme learning machine is expressed as:
[0040] Y u =f o (W o ·H u +b o )·η o
[0041] Where Y u represents the nonlinear mapping result of the output layer of the extreme learning machine; f o () represents the output layer activation function; W o represents the output layer weight of the extreme learning machine; b o represents the output layer bias term of the extreme learning machine; η o Represents the adaptive adjustment factor of the output layer of the extreme learning machine.
[0042] The present invention provides a method for monitoring the health status of periodontal tissues based on oral detection indicators, which has the following beneficial effects:
[0043] 1. The feature redundancy problem of oral examination data is solved by the entropy rate extreme learning machine algorithm. The response mode of neurons is dynamically adjusted by using the neuron entropy rate, which reduces the feature redundancy in oral examination data and enhances data discrimination. The present invention can dynamically adjust the sensitivity of the model to oral examination data, thereby improving the classification effect.
[0044] 2. Optimize the weight initialization process through the information density adjustment mechanism. The information density adjustment factor is used in the weight initialization process of the extreme learning machine. This factor is based not only on the data scale and distribution, but also takes into account the local characteristics of the data, further improving the model's adaptability and convergence speed to different features. The information density adjustment mechanism can effectively handle sparse features in high-dimensional data and optimize weight initialization during training.
[0045] 3. Avoid local optimal solutions and improve model robustness through a dynamic random jump mechanism. The dynamic random jump mechanism is used to update weights and biases, avoiding local optimal solutions that may occur in traditional gradient descent methods. It is especially effective when the data noise is large, ensuring that the model can generalize better and improving the stability and adaptability of the model. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 A comparison chart of the classification accuracy of different algorithms provided by the embodiments of the present invention;
[0047] Figure 2 A comparison chart of training time for different algorithms provided by embodiments of the present invention;
[0048] Figure 3 A schematic diagram showing the relationship between feature redundancy and classification accuracy of different algorithms provided in an embodiment of the present invention;
[0049] Figure 4 A comparison chart of model stability of different algorithms provided by embodiments of the present invention. DETAILED DESCRIPTION
[0050] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0051] It is understood that the terms "first," "second," etc., used herein may be used to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script without departing from the scope of this application.
[0052] An embodiment of the present invention provides a method for monitoring the health status of periodontal tissues based on oral detection indicators, the method comprising:
[0053] Initialize the weights, bias parameters, and information density adjustment items of the extreme learning machine, and construct the weight expression of the extreme learning machine.
[0054] In this step, the weights and bias parameters of the extreme learning machine are initialized, and the information density adjustment item is used so that the weights of the extreme learning machine depend not only on the scale and distribution of the input oral detection data, but also on the information density of the oral detection data, thereby enhancing the adaptability to the input features and the convergence speed. The features of oral detection data are complex, diverse, and high-dimensional. Directly using these data for classification is prone to feature redundancy and slow training speed. By adopting the information density adjustment factor, the algorithm can consider the distribution characteristics of data features when initializing the weights, so that the weights can better adapt to the diversity of the input data, thereby improving the convergence speed and classification accuracy of the model. When the data contains relatively sparse features, the adjustment of the weights can effectively reflect the local characteristics of the data, thereby enhancing the ability to handle abnormal data, which is expressed as:
[0055]
[0056] Where W u is the weight of the extreme learning machine; ~ means obeying a specific distribution; n u is the feature number of oral detection data input to the extreme learning machine; α ur is the weight scaling factor, which controls the amplitude of weight initialization; Indicates that the mean is 0 and the variance is σ ua 2 Normal distribution; represents normal distribution; σ ua 2 is the initialization variance of the extreme learning machine weight parameters; u represents the information density adjustment factor based on the input oral detection data characteristics. Preferably, α ur Set to 0.1.
[0057] The information density adjustment is achieved based on the normalized value of the input feature and random perturbation. The calculation method of the information density adjustment factor is expressed as:
[0058]
[0059] Where, X c,i represents the i-th eigenvalue of the oral cavity detection data input by the extreme learning machine; X c is the oral cavity detection data input to the extreme learning machine; || || is the L2 norm; τ uAn exponential factor to control the impact of information density; is the initialization variance of the information density adjustment factor; i is a positive integer. Preferably, σ3 is set to 0.01, τ u Set to 2.
[0060] Initialize the bias of the extreme learning machine, expressed as:
[0061]
[0062] Where b u is the bias of the extreme learning machine; γ uce is the bias scaling factor of the extreme learning machine. Preferably, γ uce Set to 0.04.
[0063] Calculate the entropy rate of each neuron in the extreme learning machine and use the entropy rate to measure the rate of change of oral detection data.
[0064] In this step, the entropy rate of each neuron in the extreme learning machine is calculated to measure the change rate of the oral examination data, and the dynamic adjustment of the neuron response pattern is achieved to improve the overall information extraction ability, reduce feature redundancy, and enhance the discrimination of oral examination data. Oral examination data often involves multiple time points and dynamic changes. The dynamic adjustment of the entropy rate can reflect the changes in data features in real time, thereby reducing feature redundancy and improving data discrimination. For example, for changes in periodontal health status, the entropy rate can accurately capture the subtle differences between healthy and diseased states, improving the sensitivity of the classification model. The calculation method is expressed as:
[0065]
[0066] Where H uea represents the entropy rate of the neurons of the extreme learning machine; p i represents the output response probability of the i-th neuron of the extreme learning machine; log() represents the natural logarithm function; m ur is the number of neurons in the extreme learning machine; λ3 is the mutual information regularization coefficient; represents the mutual information calculation function; H u is the hidden layer output of the extreme learning machine. Preferably, λ3 is set to 0.2.
[0067] The weight connection is optimized according to the entropy rate, and the input weight is adjusted by gradient descent. The calculation method is expressed as:
[0068]
[0069] Where, represents the weight of the extreme learning machine after entropy rate constraint optimization; H u is the hidden layer output of the extreme learning machine; is the symbol of the partial derivative.
[0070] The mutual information calculation is realized based on the joint distribution and marginal distribution of the input oral detection data and the hidden layer output, which is expressed as:
[0071]
[0072] Where, p(X c ,H u ) represents the joint probability distribution of the input oral detection data and the hidden layer output of the extreme learning machine; p(X c ) represents the marginal probability distribution of the oral cavity detection data input to the extreme learning machine; p(H u ) represents the marginal probability distribution of the hidden layer output of the extreme learning machine.
[0073] During each training iteration, weights and bias updates are adjusted through a dynamic random jump mechanism.
[0074] In this step, during each training iteration, the weight and bias updates are adjusted through a dynamic random jump mechanism to solve the problem of falling into a local optimal solution during the gradient descent process and enhance the exploratory and adaptable nature of the update strategy. For example, in the case of large data noise, the jump mechanism can avoid overfitting and ensure that the model generalizes better to unknown data. The calculation method is expressed as: ,
[0075] Where, Indicates the adjustment amount of the extreme learning machine weight for each jump; is the jump scale factor; is the jump factor strength; represents the jump initialization variance parameter, preferably, Set to 0.01.
[0076] Furthermore, the weight update is achieved based on jump adjustment, and the calculation method is expressed as:
[0077]
[0078] Where, Represents the updated extreme learning machine weights.
[0079] Adaptive jump strength is used to dynamically adjust the jump factor strength according to the current loss function gradient to prevent excessive perturbation of model convergence. The calculation method of the jump factor strength is expressed as: ,
[0080] Where η u represents the basic jump strength; γ1 is the adaptive adjustment factor; represents the gradient of the extreme learning machine's loss function with respect to the weights. Preferably, η u Set γ to 0.2 and γ1 to 0.5.
[0081] The hidden layer of the extreme learning machine contains a feature reconstruction module, which reconstructs the features by analyzing the correlation between the features in the training oral detection data and optimizes the mapping space.
[0082] In this step, a feature reconstruction module is added to the hidden layer of the extreme learning machine. By analyzing the correlation between the features in the training oral examination data, feature reconstruction and mapping space optimization are achieved to solve the problem of high-dimensional oral examination data and feature redundancy, and improve feature expression ability and classification accuracy. In traditional classification models, redundant features often reduce the efficiency of the model, while feature reconstruction reduces the interference of redundant information and enhances classification accuracy by mapping to a new feature space. For example, in the periodontal health assessment task, different oral examination data features (such as gum depth, tartar distribution, etc.) may affect each other. The features optimized by the reconstruction module can more clearly distinguish between healthy and diseased states. The calculation method is expressed as:
[0083]
[0084] Where, Represents the features reconstructed by the feature reconstruction module; X c represents the input oral cavity detection data of the feature reconstruction module; M u Represents the feature mapping matrix, which determines how the input features are mapped to the new feature space.
[0085] The feature mapping matrix is optimized based on the reconstruction error gradient. The update method of the feature mapping matrix is expressed as:
[0086]
[0087] Where, Represents the updated feature mapping matrix; λ2 is the learning rate of mapping optimization; Represents the reconstruction error E u About the feature mapping matrix M u gradient; Denotes the noise error term. Preferably, λ2 is set to 0.03.
[0088] The feature reconstruction error is measured based on the difference in oral detection data, and the calculation method is expressed as:
[0089]
[0090] Where, E u Represents the reconstruction error of the feature reconstruction module; n uyrThe number of samples input to the extreme learning machine for the current batch.
[0091] In the hidden layer output stage of the extreme learning machine, the connection weights between neurons are adjusted through back propagation, and the hidden layer output of the extreme learning machine is calculated.
[0092] In this step, at the hidden layer output stage of the extreme learning machine, the connection weights between neurons are adjusted through back propagation, and the hidden layer output of the extreme learning machine is calculated, which is expressed as:
[0093]
[0094] Where H u is the hidden layer output of the extreme learning machine; Sig() is the Sigmoid activation function; represents the updated extreme learning machine weight; Represents the features reconstructed by the feature reconstruction module.
[0095] In the output layer of the extreme learning machine, a nonlinear activation function is used and iterations are repeated until the stopping condition is met. The health status of periodontal tissue is monitored through the trained extreme learning machine.
[0096] In this step, the output layer of the extreme learning machine uses nonlinear output regulation to enhance the ability to distinguish different categories. The classification task of oral examination data has strong nonlinear characteristics. Therefore, the nonlinear activation function of the output layer can effectively improve the model's ability to handle complex classification tasks. In periodontal health assessment, the model needs to distinguish different health states based on the input data to help the model make more accurate classification decisions. The calculation method is expressed as:
[0097] Y u =f o (W o ·H u +b o )·η o
[0098] Where Y u represents the nonlinear mapping result of the output layer of the extreme learning machine; f o () represents the output layer activation function; W o represents the output layer weight of the extreme learning machine; b o represents the output layer bias term of the extreme learning machine; η o Represents the adaptive adjustment factor of the output layer of the extreme learning machine. Preferably, the output layer activation function adopts the Softmax function.
[0099] Furthermore, the output adjustment factor is dynamically adjusted according to the current gradient, and the calculation method is expressed as:
[0100]
[0101] Where, γ2 is the basic coefficient of the extreme learning machine output adjustment; β2 is the gradient influence factor; Denotes the loss function L u About weight W u gradient; represents the gradient of the extreme learning machine's loss function with respect to the weights. Preferably, γ2 is set to 0.2 and β2 is set to 0.5.
[0102] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.
[0103] Effect verification:
[0104] like Figure 1 As shown, in order to verify the effectiveness of the technical method of the present invention, the comparative bar chart of algorithm accuracy shows the performance of different classification algorithms (including entropy rate extreme learning machine, SVM, random forest and decision tree) in the periodontal tissue health status classification task. The experimental results show that the classification accuracy of the entropy rate extreme learning machine is higher than that of other conventional algorithms, showing its superiority in complex oral detection data. By dynamically adjusting the response mode of neurons and reducing feature redundancy, this technology effectively improves the discrimination of oral detection data, especially when processing oral data with diversity and high dimensionality, its accuracy performance is more outstanding.
[0105] like Figure 2 As shown in the figure, the training time comparison line chart shows the changes in training time of different algorithms when processing large-scale data. As the amount of data increases, the training time of the entropy rate extreme learning machine shows a more gradual growth trend than other traditional algorithms, indicating that the algorithm has higher training efficiency when processing large data sets. It can effectively reduce feature redundancy and avoid local optimal solutions through a dynamic jump mechanism during the algorithm training process, ensuring the rapid convergence of the model.
[0106] like Figure 3 As shown in the figure, the effect of feature redundancy on classification accuracy is demonstrated through heat maps, especially the performance of entropy rate extreme learning machine and other algorithms in dealing with feature redundancy is compared. The results show that when feature redundancy is high, entropy rate extreme learning machine can maintain high classification accuracy, while the accuracy of other algorithms decreases significantly, highlighting the advantages of this technology in reducing redundant features and optimizing feature representation, and can more effectively extract useful information from high-dimensional complex data.
[0107] like Figure 4 As shown in the figure, the scatter plot shows the stability of the model under different noise levels, especially in the case of data imbalance and feature loss. The experimental results show that the entropy rate extreme learning machine is more robust in dealing with noise and data disturbances, and the classification accuracy decreases less. In contrast, the accuracy of traditional SVM and K-nearest neighbor algorithms drops rapidly when the noise increases, highlighting the unique advantages of this technology in improving the robustness and adaptability of the model. Through the adaptive jump mechanism and dynamic adjustment of weight bias, the entropy rate extreme learning machine effectively avoids local optimal solutions and ensures the high stability and high accuracy of the model.
[0108] By dynamically adjusting the entropy rate, the present invention can more accurately capture the subtle differences between healthy and diseased states, reduce the interference of irrelevant features, and improve the accuracy of oral examination data classification;
[0109] The information density adjustment factor greatly improves the adaptability of weight initialization to the input feature distribution, accelerates the training process, and avoids low training efficiency caused by uneven distribution of data features;
[0110] The dynamic random jump mechanism effectively avoids the problem of traditional gradient descent methods falling into local optimal solutions during training, allowing the model to maintain high accuracy in the face of noise and disturbances, reducing the risk of overfitting.
[0111] The feature reconstruction module optimizes the mapping of input data, reducing the negative impact of redundant features on the model. This module can effectively reduce redundancy in high-dimensional oral examination data and improve the classification ability of the model.
[0112] The nonlinear output regulation mechanism adaptively adjusts the activation function, enabling the model to more accurately distinguish different health states and provide higher classification accuracy, especially in processing complex and nonlinear feature data.
[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A method for monitoring the health status of periodontal tissue based on oral detection indicators, characterized in that: The method comprises: Initialize the extreme learning machine's weights, bias parameters, and information density adjustment items, and construct the extreme learning machine's weight expression; Calculate the entropy rate of each neuron in the extreme learning machine and use the entropy rate to measure the rate of change of oral detection data; During each training iteration, the weights and biases are updated and adjusted through a dynamic random jump mechanism; The hidden layer of the extreme learning machine contains a feature reconstruction module, which reconstructs features by analyzing the correlation between each feature in the training oral examination data and optimizes the mapping space; In the hidden layer output stage of the extreme learning machine, the connection weights between neurons are adjusted through back propagation, and the hidden layer output of the extreme learning machine is calculated; In the output layer of the extreme learning machine, a nonlinear activation function is used and iterations are repeated until the stopping condition is met. The health status of periodontal tissue is monitored through the trained extreme learning machine.
2. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 1, characterized in that: The weight of the extreme learning machine is expressed as: Where W u is the weight of the extreme learning machine; ~ means obeying a specific distribution; n u is the feature number of oral detection data input to the extreme learning machine; α ur is the weight scaling factor, which is used to control the amplitude of weight initialization; Indicates that the mean is 0 and the variance is σ ua 2 Normal distribution; represents normal distribution; σ ua 2 is the initialization variance of the extreme learning machine weight parameters; u Represents the information density adjustment factor based on the characteristics of the input oral detection data.
3. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 2, characterized in that: The information density adjustment factor is expressed as: Where, X c,i represents the i-th eigenvalue of the oral cavity detection data input by the extreme learning machine; X c is the oral cavity detection data input to the extreme learning machine; |||| is the L2 norm; τ u An exponential factor to control the impact of information density; is the initialization variance of the information density adjustment factor; i is a positive integer.
4. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 1, characterized in that: The entropy rate of a neuron is expressed as: Where H uea represents the entropy rate of the neurons of the extreme learning machine; p i represents the output response probability of the i-th neuron of the extreme learning machine; log() represents the natural logarithm function; m ur is the number of neurons in the extreme learning machine; λ3 is the mutual information regularization coefficient; I() represents the mutual information calculation function; H u is the hidden layer output of the extreme learning machine.
5. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 1, characterized in that: In the step of adjusting the weight and bias updates through the dynamic random jump mechanism, the adjustment amount of the extreme learning machine weight is expressed as: In the formula, ΔW u Represents the adjustment amount of the extreme learning machine weight for each jump; β1 is the jump scale factor; ∈ u is the jump factor strength; represents jump initialization variance parameter; The update process of the extreme learning machine weight is expressed as: Where, Represents the updated extreme learning machine weights.
6. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 5, characterized in that: In the step of adjusting the weight and bias updates through the dynamic random jump mechanism, adaptive jump strength is used to dynamically adjust the jump factor strength according to the current loss function gradient. The calculation method of the jump factor strength is expressed as: Where η u represents the basic jump strength; γ1 is the adaptive adjustment factor; Represents the gradient of the extreme learning machine's loss function with respect to the weights.
7. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 1, characterized in that: The hidden layer of the extreme learning machine contains a feature reconstruction module. By analyzing the correlation between the features in the training oral examination data, the features are reconstructed and the mapping space is optimized. The feature representation after feature reconstruction is: Where, Represents the features reconstructed by the feature reconstruction module; X c represents the input oral cavity detection data of the feature reconstruction module; M u Represents the feature mapping matrix, which determines how the input features are mapped to the new feature space.
8. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 7, characterized in that: The feature mapping matrix is optimized based on the reconstruction error gradient. The update method of the feature mapping matrix is expressed as: Where, Represents the updated feature map matrix; λ2 is the learning rate for mapping optimization; Represents the reconstruction error E u About the feature mapping matrix M u gradient; represents the noise error term.
9. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 1, characterized in that: In the hidden layer output stage of the extreme learning machine, the connection weights between neurons are adjusted through back propagation, and the hidden layer output of the extreme learning machine is calculated. The hidden layer output of the extreme learning machine is expressed as: Where H u is the hidden layer output of the extreme learning machine; Sig() is the Sigmoid activation function; represents the updated extreme learning machine weight; Represents the features reconstructed by the feature reconstruction module.
10. The method for monitoring periodontal tissue health status based on oral detection indicators according to claim 1, characterized in that: In the output layer of the extreme learning machine, when a nonlinear activation function is used, the nonlinear mapping result of the extreme learning machine output layer is expressed as: Y u =f o (W o ·H u +b o )·η o Where Y u represents the nonlinear mapping result of the output layer of the extreme learning machine; f o () represents the output layer activation function; W o represents the output layer weight of the extreme learning machine; b o represents the output layer bias term of the extreme learning machine; η o Represents the adaptive adjustment factor of the output layer of the extreme learning machine.