Dynamic regulation method and system of tooth whitening instrument based on multi-modal perception

Abstract

The application discloses a dynamic regulation method and system of a tooth whitening instrument based on multi-modal perception, and the method comprises the following steps: S1: collecting tooth images, tooth surface temperatures and user comfort rating data through multi-modal sensors; S2: calculating a coloring index, a temperature state and a user comfort state based on the tooth images, the tooth surface temperatures and the user comfort rating data; S3: generating a comprehensive state index based on the three states; S4: determining whether the comprehensive state index exceeds a preset threshold, if yes, executing step S5, otherwise maintaining the current whitening parameters; S5: dynamically solving whitening power parameters at the next moment based on a target function; S6: controlling a light source module to output according to the solved power parameters; and S7: determining whether to end the process or return to step S1 according to the whitening effect. Through real-time fusion of multi-modal data such as vision, temperature and user feedback, the application adaptively adjusts the whitening parameters, and realizes dynamic balance of the whitening efficiency, safety and comfort.

Description

Technical Field

[0001] This invention belongs to the field of oral care technology, specifically relating to a personalized teeth whitening instrument with dynamic control device and algorithm based on multimodal perception. Background Technology

[0002] Most existing teeth whitening devices operate using preset timers or fixed power output modes, such as activating a whitening agent through cold light to achieve a whitening effect. Some high-end devices introduce user selection modules (e.g., "7-day whitening" or "30-day whitening"), allowing the reception of desired results data via an radio frequency module. However, such adjustments are still based on static presets and cannot be dynamically optimized according to real-time changes in the oral environment. While existing technologies attempt to detect tooth hue using color sensors or control light source output using pulse width modulation (PWM) circuits, their control logic remains limited to single-modal data (e.g., color or pigmentation), lacking comprehensive perception and response to multi-dimensional information such as temperature and user subjective feelings. For example, some solutions only detect tooth hue using color sensors and then control light source output via PWM circuits, but fail to consider the impact of oral temperature changes on tooth enamel and cannot capture user subjective feelings (e.g., stinging sensations), thus making it difficult to achieve an effective balance between whitening effect and safety.

[0003] Furthermore, existing teeth whitening devices have significant shortcomings in real-time monitoring and interaction. For example, when there is a risk of overheating during whitening agent irradiation, the device lacks an effective temperature feedback mechanism to prevent damage to the gums or enamel; simultaneously, user feedback on irritation or pain cannot be captured by the system in real time and used for parameter adjustment. Although some studies have proposed artificial intelligence-based dental aesthetic analysis systems (such as image feature extraction schemes using CNN-LSTM networks), these rely on historical databases rather than real-time data streams and do not involve photothermal coupling control or dynamic closed-loop regulation. These deficiencies make it difficult for existing teeth whitening devices to achieve a balance between safety, comfort, and whitening effect. Some recent designs attempt to introduce the concept of zoned whitening (such as achieving zone customization through independently controlled multi-LED lights), but their control signals still rely on user presets and fail to dynamically adjust based on real-time sensing data (such as temperature and user feedback), resulting in a lack of adaptive capabilities in the whitening process. Summary of the Invention

[0004] A brief overview of embodiments of the invention is provided below to provide a basic understanding of certain aspects of the invention. It should be understood that this overview is not an exhaustive summary of the invention. It is not intended to identify key or essential parts of the invention, nor is it intended to limit the scope of the invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0005] To address the aforementioned technical issues, this invention achieves a dynamic balance between whitening efficiency, safety, and comfort by real-time fusion of multimodal data, including visual, temperature, and user feedback, and adaptively adjusting whitening parameters.

[0006] Specifically, according to one aspect of this application, a dynamic control method for a teeth whitening device based on multimodal perception is provided, comprising: S1: Acquires tooth images, tooth surface temperature, and user comfort rating data through multimodal sensors; S2: Calculate the color index, temperature status, and user comfort status based on tooth images, tooth surface temperature, and user comfort score data; S3: Generate a comprehensive state index based on the weighted coefficient fusion of color index, temperature status and user comfort status; S4: Determine whether the comprehensive status index exceeds the preset threshold. If yes, proceed to step S5; otherwise, maintain the current whitening parameters. S5: Dynamically solve the whitening power parameters for the next time step based on the objective function; the objective function is constructed based on whitening effect, thermal safety, and comfort. S6: Control the output of the light source module according to the solved power parameters; S7: Evaluate the whitening effect. If the pigmentation index is lower than the target value or the maximum duration is reached, end the process; otherwise, return to step S1.

[0007] Furthermore, the tinting index is obtained through the following process: converting the acquired tooth image to the LAB color space, calculating the color difference between the tooth region and the standard whiteness, and normalizing it to obtain the tinting index. .

[0008] Furthermore, the temperature state The predicted temperature at time t Oral baseline temperature The difference, the predicted temperature A discretized model based on the law of conservation of energy, according to real-time temperature. Illumination power The photothermal conversion coefficient kl and the heat dissipation coefficient ηs were calculated.

[0009] Furthermore, the predicted temperature at time t is expressed as follows: Where kl is the photothermal conversion coefficient and ηs is the heat dissipation coefficient. This is the current illumination power. For time step.

[0010] Furthermore, the user comfort state By scoring users' real-time comfort levels The result is obtained after normalization.

[0011] Furthermore, the comprehensive state index Calculated using the following formula: ; in, , , These are weighting coefficients, calibrated experimentally (default). =0.5, =0.3, =0.2), reflecting the bias towards different modal data; The color index at time t; The difference between the predicted temperature at time t and the baseline oral temperature; Let t represent the user's comfort level at time t.

[0012] Furthermore, in step S5, the objective function is: ; in, , , To optimize weights, dynamic solution is used under constraints. and The following steps are performed to obtain the optimal power solution. That is, the most beautiful white power; Furthermore, the optimized weights , , It can be updated online using gradient descent, and the update process is based on a loss function that includes the whitening effect error and the user comfort penalty. .

[0013] Furthermore, the optimized weights are updated based on gradient descent, specifically by dynamically balancing the weights by calculating the gradient of the loss function in real time. , , The expression is as follows: i=1,2,3 The weight value at time t+1. Let be the weight value at time t, and η be the learning rate, used to control the step size for each update. Let η be the gradient of the loss function with respect to the weights, i.e., the partial derivative. This mechanism allows the system to automatically adjust the weights based on user feedback after each use (such as a decrease in comfort rating), improving personalized adaptability. The learning rate η adaptively decays over time to ensure system convergence.

[0014] Furthermore, the loss function The expression is as follows: ; in, The new pigmentation index after applying whitening parameters is either a predicted or measured value. The target value for the coloring index can be set to, for example, 0.08; It is a real-time user comfort rating. The target value for comfort can be set, for example, to 0.0132; The comfort penalty weight is set to 0.5 by default, which is used to balance the importance of whitening effect and user experience. Right now .

[0015] According to a second aspect of this application, a dynamic control system for a teeth whitening device based on multimodal perception is provided, comprising: The main control module is used to execute the control algorithm; Image sensor used to acquire images of teeth and calculate the staining index; Temperature sensor used to monitor tooth surface temperature; The user feedback module is used to receive user comfort ratings. The light source module is used to provide whitening light. The main control module is configured to: integrate the coloring index, the temperature state obtained based on the temperature prediction model, and the user comfort state to generate a comprehensive state index, and dynamically solve the objective function based on the comprehensive state index to optimize the whitening power parameters.

[0016] According to a third aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the dynamic control method described above.

[0017] The present invention employs the above-described scheme to realize a dynamic control scheme for a teeth whitening device based on multimodal perception, which has the following advantages compared with the prior art: 1. Multimodal perception enhances personalized accuracy: By fusing image, temperature, and user feedback data, it overcomes the limitations of single-modality sensing and dynamically adapts to individual differences (such as differences in pigmentation and heat resistance). Compared to traditional single-color sensor solutions, the recognition accuracy is improved by approximately 50%, and the consistency of whitening effects is significantly enhanced.

[0018] 2. Real-time closed-loop control ensures safety: Overheating damage is prevented through real-time closed-loop control and heat accumulation prediction model. Combined with comfort constraints, a balance between whitening efficiency and safety is achieved, which can effectively reduce temperature over-limit events and user discomfort.

[0019] 3. Strong algorithm adaptability: weight coefficient ( , , (etc.) can learn online based on historical data and adapt to changes in oral condition during long-term use.

[0020] 4. Excellent scalability: The framework supports the integration of more modalities (such as saliva pH monitoring) and reserves interfaces for future upgrades. It employs a feature-level fusion method, facilitating the rapid access and processing of new sensor data.

[0021] In summary, this invention improves the whitening effect of a teeth whitening device through multimodal perception, introduces a heat accumulation prediction model into the temperature control of the device, and applies and improves the gradient descent online update method to balance the whitening efficiency and safety of the device. This results in a dynamic control scheme for a teeth whitening device based on multimodal perception, which has great practicality. Attached Figure Description

[0022] The present invention can be better understood by referring to the description given below in conjunction with the accompanying drawings, in which the same or similar reference numerals are used throughout the drawings to denote the same or similar parts. These drawings, together with the following detailed description, are incorporated in and form part of this specification, and are used to further illustrate preferred embodiments of the invention and explain the principles and advantages of the invention. In the drawings: Figure 1 This is a flowchart of the dynamic control method of the teeth whitening device according to an embodiment of the present invention. Detailed Implementation

[0023] Embodiments of the present invention will now be described with reference to the accompanying drawings. Elements and features described in one drawing or embodiment of the invention may be combined with elements and features shown in one or more other drawings or embodiments. It should be noted that, for clarity, representations and descriptions of components and processes unrelated to the present invention and known to those skilled in the art have been omitted from the drawings and description.

[0024] Example 1 This embodiment provides a dynamic adjustment algorithm based on multimodal perception, which adaptively adjusts whitening parameters by fusing visual, temperature, and user feedback data in real time. The technical solution is described in detail below, step by step.

[0025] Step 1: Multimodal Sensing and Data Acquisition The teeth whitening device system collects data in real time through the following sensors: (1) Image sensor: An embedded miniature camera (e.g., 640×480 resolution, 5fps sampling frequency) is used to capture images of the tooth surface. The L (luminance), a (green-red), and b (blue-yellow) channel values ​​of the tooth area are extracted through LAB color space conversion, and the chromaticity index is calculated.

[0026] (2) Temperature sensor: The DS18B20 digital temperature sensor (accuracy ±0.1℃, sampling frequency 10Hz) is used to monitor the tooth surface temperature in real time. The baseline temperature is set to the normal oral temperature of 35℃.

[0027] (3) User feedback module: Through the interactive interface on the handheld terminal (such as mobile phone or tablet), the user can input a comfort score (1-5 points) in real time. The score is transmitted to the main control module via Bluetooth. The score is inversely proportional to the comfort level. The higher the score, the less comfortable the user is.

[0028] Step 2: Multimodal Data Fusion Strategy Inspired by machine synesthesia frameworks, this invention employs a feature-level intermediate fusion method to map heterogeneous data into a unified state vector. The multimodal data vector at time t is defined as follows: , , . The tooth staining index is extracted through image analysis, see step three; The temperature state at time t is equal to the predicted temperature at time t. Oral baseline temperature The difference, i.e. (Unit: °C), Predicted Temperature The calculation process is shown in step four; The baseline temperature is 35°C, which is the normal oral temperature. User comfort level is normalized to the range of 0-1. , Rate the user's comfort level at the current time.

[0029] The comprehensive state index is obtained through weighted fusion: ; Parameter description: , , To integrate the weighting coefficients of tinting index, temperature condition, and comfort condition, experimental calibration was performed (default). =0.5, =0.3, =0.2), reflecting the bias towards different modal data; Color difference calculation based on image LAB space; Reflects the risk of heat accumulation; Quantifying subjective feelings about the reaction.

[0030] Step 3: Design of Color Degree Quantization Algorithm Tooth discoloration index Quantization is a multi-step image processing and analysis process. Its core idea is to convert the tooth image from the universal RGB color space to the Lab color space, which is more in line with human visual perception, and to quantify the degree of coloring by calculating the color difference with the ideal whiteness.

[0031] First, image acquisition and preprocessing are performed. Under the illumination of a standard D65 light source (a standard light source simulating sunlight), RGB images of the teeth are acquired through a miniature camera built into the teeth whitening device. The acquired raw images undergo preprocessing steps such as correcting the inherent radial and tangential distortion of the camera lens, white balance adjustments, and ROI (region of interest) selection to eliminate interference.

[0032] Next, a color space conversion is performed, transforming the preprocessed RGB image to the LAB color space. The conversion process first converts RGB to the XYZ tristimulus space, and then to the Lab space (using a general formula). The formula is as follows: First, convert it to XYZ space: X=0.4124564×R+0.3575761×G+0.1804375×B; Y=0.2126729×R+0.7151522×G+0.0721750×B; Z=0.0193339×R+0.1191920×G+0.9503041×B; R, G, and B represent the values ​​for the three primary colors: red, green, and blue, respectively.

[0033] Then convert it to Lab space: ; ; ; ; ; This piecewise function is designed to maintain linearity at low brightness levels and avoid calculating singularities.

[0034] Calculate the tooth area and standard whiteness (reference value) =95, =0, Color difference of 0 (=0): ; Colorimetric index normalization: ; Parameter description: , , The LAB channel mean of the tooth image is extracted by an image segmentation algorithm (based on U-Net network) after converting the RGB image to the LAB color space using the above formula; The maximum color difference threshold is set to 35 based on clinical experience. Teeth stained >35 are typically considered severely stained (e.g., tetracycline-stained teeth), requiring special treatment or multiple whitening sessions. Normalization ensures... It is a dimensionless exponent, where 1 represents the most severe coloration. Considering work efficiency, it is generally not pursued. Equal to 0, when =0.1 indicates that the target whiteness has been achieved.

[0035] Step 4: Design of Heat Accumulation Prediction Model By introducing a heat accumulation prediction model, instead of simply relying on real-time temperature measurement, the control system is upgraded from a passive response to a proactive prevention approach. Essentially, this is based on an understanding of physical processes to anticipate the future and make better decisions. This allows the teeth whitening device to dynamically maintain a balance between whitening efficiency and safety boundaries, providing effective light power as continuously as possible without causing burns, avoiding frequent interruptions caused by simple threshold cutoffs, thereby improving the overall whitening effect and user experience.

[0036] According to the law of conservation of energy, the rate of temperature change of teeth depends on the light power. The net difference between the generated heat input and the ambient heat dissipation power; after discretization (time step). =1s), the formula for predicting the temperature is expressed as follows: ; Parameter description: kl=0.02 is the photothermal conversion coefficient, which is obtained through calibration experiments. That is, in the laboratory, the temperature rise curves of the tooth surface under different powers P are recorded, and the relationship between the temperature rise rate and the power is fitted by linear regression; ηs=0.1 is the heat dissipation coefficient, which depends on the humidity of the oral environment and the saliva flow rate. The illumination power (adjustable from 0-100%) is the illumination power applied at time t. The tooth surface temperature is the real-time output of the temperature sensor. Baseline temperature (normal oral temperature 35℃).

[0037] Step 5: Dynamic Adjustment Strategy Design an objective function to balance skin whitening efficiency and safety: ; Constraints: ≤ ≤ ( =20%, =80%) ≤ (Safety threshold 38℃).

[0038] Parameter description: , , To optimize weights (default) =0.6, =0.25, =0.15), updated online using gradient descent; The smaller the tinting index (the whiter), the larger this item is; The larger this value is, the lower the actual temperature is compared to the safety threshold, and the more compliant it is with safety requirements. A higher comfort level value indicates greater discomfort. The optimal power solution is obtained by solving the above constrained optimization problem; that is, under the constraints, an optimal power solution is found that makes the objective function... maximum.

[0039] Step Six: Design the calculation process for optimizing the online update of weights using gradient descent. This step elaborates on the online update of the gradient descent method in the dynamic adjustment strategy of step five, and can be divided into four parts: defining the loss function, gradient descent design, gradient calculation, and online update process.

[0040] 1. Define the loss function A loss function designed to be minimized is defined. It reflects the gap between the current whitening effect and the ideal state. This loss function consists of two parts: effect error and comfort penalty. The effect error refers to the pigmentation index after whitening. The gap from the ideal target (0.1). The comfort penalty is the user's real-time comfort score. The penalty for a score that is too low (indicating discomfort).

[0041] ; in, The new pigmentation index after applying whitening parameters; either a predicted or measured value. The target value for the coloring index can be set to, for example, 0.08; It is a real-time user comfort score, normalized to 0-1. The target value for comfort can be set as a small non-zero value (0.0132) based on engineering experience. This represents the pursuit of ultimate comfort (approaching 0) and may also be used to avoid singularity problems in numerical calculations in actual engineering calculations. The comfort penalty weight is set to a default value of 0.5 to balance the importance of whitening effect and user experience.

[0042] 2. Gradient Descent Design That is, along the opposite direction of the gradient (derivative) of the loss function, the parameters are adjusted in small steps (learning rate) to gradually approach the minimum value of the loss function. The gradient is... , which indicates the direction in which the loss function changes the fastest along the parameter axis at the current parameter value point.

[0043] The parameter update formula is designed as follows: ; in, The weight value at time t+1. Let be the weight value at time t. η is the learning rate, which controls the step size of each update, and is generally set to 0.01. A learning rate that is too small will lead to slow convergence, while a learning rate that is too large will lead to oscillations or even divergence. This is the gradient of the loss function with respect to the weights, i.e., the partial derivative.

[0044] 3. Gradient calculation using the chain rule. To update the weights in the objective function below For example, this demonstrates how to calculate the gradient. It requires calculating... According to the chain rule, gradient calculation can be decomposed into: ; The first item: = This can be obtained by directly differentiating the loss function.

[0045] Second item: Describes whitening power How does the change in power affect the change in the pigmentation index I(t+1)? This relationship is difficult to describe with a precise physical model, so a simple linear approximation is used: as power increases, the whitening effect is enhanced, and the pigmentation index decreases. It can be approximated as a negative constant, such as... ≈-0.005.

[0046] Third item: The objective function J(P) is a pair The partial derivative in the optimal solution The value at that point. According to the objective function formula... ,so In the optimal solution The place can be considered Related to this, to simplify calculations, its sign or a proportional value can be taken, such as... , It is a special symbol in mathematics used to represent a direct proportional relationship between two quantities.

[0047] Combining these three factors yields the gradient estimate of the weights. This estimate will guide... The direction of updates, , The same process is used for the calculation.

[0048] (4) Online update process The weight update process is a continuous cycle of fine-tuning during each whitening treatment: Initial values: At time t=0, set initial weight values; all weights start from a set of empirical default values ​​(e.g., ...). =0.6, =0.25, =0.15).

[0049] Execution and control: Based on the current weights, solve the objective function to obtain the optimal white parameters for this round. And execute.

[0050] Feedback collection: The pigmentation index will be measured again shortly after the whitening effect (at time t+1 of the next sampling period). (Or use predicted values), and obtain user comfort scores. .

[0051] Calculate the loss and gradient: Calculate the loss function according to the above loss function formula. and gradients with respect to each weight .

[0052] Update weights: Apply the gradient descent formula to update all weights.

[0053] i=1,2,3; That is, the loss function mentioned above. .

[0054] Loop: The updated weights are used for the next round of whitening parameter calculations, and this process is repeated.

[0055] Through this online learning mechanism, the teeth whitening device can gradually "understand" the characteristics of the current user: for example, if a user is particularly sensitive to light and heat (and has a consistently low comfort score), the system will automatically reduce... (Effect weighting) and increase (Comfort weight) Pursues whitening effect while ensuring basic safety and comfort, truly achieving highly personalized dynamic control.

[0056] Step 7: Overall Workflow Description (1) Initialization: Set initial parameters ( =35℃, =40%).

[0057] (2) Data acquisition: Real-time acquisition of images, temperature and user ratings.

[0058] (3) State calculation: Calculated according to the comprehensive state index formula. ,judge Does it exceed the preset threshold? > If the threshold is 0.7, then parameter optimization and adjustment are triggered, and the process jumps to step (4); otherwise, the current whitening parameters are maintained and the process proceeds to step (5).

[0059] (4) Parameter optimization: Solve the objective function of the formula and update .

[0060] (5) Execution control: Adjust the output of the cold light source through PWM waveform.

[0061] (6) Termination of Judgment: Evaluate the whitening effect. When the pigmentation index is lower than the target value (e.g., ...), the whitening effect is terminated. Stop when <0.1) or when the maximum duration (15 minutes) is reached; otherwise, return to step (2) to continue data collection and dynamic control.

[0062] Example 2 This embodiment provides a dynamic control system based on multimodal sensing, including: Main control module: Used to receive data from various sensors and execute control algorithms; Visual sensing module: includes an embedded miniature camera responsible for acquiring images of the tooth surface; Temperature sensing module: includes a digital temperature sensor responsible for monitoring the temperature of the tooth surface; User interaction module: including Bluetooth communication module and handheld terminal, responsible for receiving user comfort ratings; Execution module: includes a blue LED light source array and a PWM drive circuit, responsible for outputting adjustable cold light irradiation.

[0063] In this embodiment, the main control module is implemented using an STM32H750VBT6 (ARM Cortex-M7 core) main control chip. The visual sensing module is implemented using an OV2640 (2-megapixel, LAB color space real-time conversion) image sensor, the temperature sensing module is implemented using a DS18B20 (accuracy ±0.1℃) temperature sensor, and the execution module includes a 460nm blue LED array (PWM dimming, power 0-100mW).

[0064] Taking a 15-minute whitening process as an example, the system workflow includes: (1) Initial stage (t=0): Acquire initial tooth images and calculate I(0) = 0.85 (indicating significant staining); Temperature T(0) = 35.0℃, user comfort S(0) = 1.0 (no discomfort before start); The initial power P(0) = 40%.

[0065] (2) Regulation phase (t=5min): Real-time data: I=0.70, T=36.5℃, S=4 (slight feeling of warmth); Calculate Z = 0.5 × 0.70 + 0.3 × 1.5 + 0.2 × 0.75 = 0.35 + 0.45 + 0.15 = 0.95 (exceeding the threshold); Solving the objective function yields P = 55%; Optimize weight coefficients , , Continuing to solve the objective function, we obtain P=54%.

[0066] (3) Termination phase (t=15min): When I=0.08 (target achieved), irradiation will automatically stop.

[0067] In practical applications, implementing the above solution in dental instruments with limited hardware resources presents significant technical challenges, primarily related to computational complexity and memory consumption. To address these challenges, this invention employs targeted measures: based on the law of conservation of energy, the continuous temperature prediction model is pre-discretized, constructing a prediction model based on real-time temperature, light power, photothermal conversion coefficient, and heat dissipation coefficient. This avoids online numerical integration, thereby reducing computational load.

[0068] This invention employs the aforementioned algorithm, collecting data through multimodal sensors (including a visual sensor for tooth coloration, a temperature sensor, and a real-time user feedback module), and using a feature-level fusion strategy to map heterogeneous data into a unified state vector. Based on this state vector, the algorithm dynamically solves for the optimal whitening parameters (such as cold light intensity, irradiation time, and cooling interval) through an objective function that includes heat accumulation prediction, coloration assessment, and comfort optimization. This invention is the first to introduce machine synesthesia into the field of oral care, achieving real-time personalized control of whitening parameters through a communication-perception integrated framework, solving the problems of static settings, lack of real-time feedback, and over-reliance on human experience in existing technologies. Compared with existing technologies, this invention has the following advantages: 1. Multimodal Perception Enhances Personalized Accuracy: By fusing image, temperature, and user feedback data, this invention overcomes the limitations of single-modal perception and dynamically adapts to individual differences (such as variations in pigmentation and heat resistance). Existing technologies often employ fixed-weighted averaging, such as S=w1×S1+w2×S2 (w1, w2 are constants), or simple logical selection, which cannot adapt to the sensitivity differences among different users and in different tooth conditions. This invention uses a feature-level intermediate fusion method to map heterogeneous data into a unified state vector. Compared to traditional single-color sensor solutions, this improves recognition accuracy and significantly enhances the consistency of whitening effects.

[0069] 2. Real-time Closed-Loop Control Ensures Safety: Existing technologies generally employ empirical threshold methods or simple linear accumulation models, neglecting the physical processes of energy transfer. This invention introduces a heat accumulation prediction model, rather than simply relying on real-time temperature measurements, upgrading the control system from passive response to proactive prevention. Essentially, it anticipates the future based on an understanding of physical processes, thereby making better decisions. This allows the teeth whitening device to dynamically maintain a balance between whitening efficiency and safety. By preventing overheating damage through the heat accumulation prediction model and combining it with comfort constraints, a balance between whitening efficiency and safety is achieved.

[0070] 3. Strong Algorithm Adaptability: Existing technologies mostly use a two-level power switching mechanism ("on-off" or "high-low"), resulting in crude control and difficulty in balancing whitening efficiency and comfort. This invention designs an objective function that balances whitening efficiency and safety, and uses online stochastic gradient descent to update key parameters. Weighting coefficients ( , , (etc.) It can learn online based on historical data and adapt to changes in oral condition over long-term use. The optimized weights are updated using gradient descent, specifically by dynamically balancing the weights by calculating the gradient of the loss function in real time. This allows the model to dynamically fine-tune its parameters based on the actual response of the user's teeth during use, achieving true personalized adaptation.

[0071] 4. Excellent scalability: The framework supports the integration of more modalities (such as saliva pH monitoring) and reserves interfaces for future upgrades. It employs a feature-level fusion method, facilitating the rapid access and processing of new sensor data.

[0072] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0073] Furthermore, the method of the present invention is not limited to being executed in the chronological order described in the specification, but may also be executed in other chronological orders, in parallel, or independently. Therefore, the execution order of the method described in this specification does not constitute a limitation on the technical scope of the present invention.

[0074] Although the invention has been disclosed above through the description of specific embodiments, it should be understood that all the embodiments and examples described above are exemplary and not restrictive. Those skilled in the art can design various modifications, improvements, or equivalents to the invention within the spirit and scope of the appended claims. These modifications, improvements, or equivalents should also be considered to be included within the protection scope of the invention.

Claims

1. A dynamic control method for a teeth whitening device based on multimodal perception, characterized in that: include: S1: Acquires tooth images, tooth surface temperature, and user comfort rating data through multimodal sensors; S2: Calculate the color index, temperature status, and user comfort status based on tooth images, tooth surface temperature, and user comfort score data; S3: Generate a comprehensive state index based on the weighted coefficient fusion of color index, temperature status and user comfort status; S4: Determine whether the comprehensive status index exceeds the preset threshold. If yes, proceed to step S5; otherwise, maintain the current whitening parameters. S5: Dynamically solve the whitening power parameters for the next time step based on the objective function; the objective function is constructed based on whitening effect, thermal safety, and comfort. S6: Control the output of the light source module according to the solved power parameters; S7: Evaluate the whitening effect. If the pigmentation index is lower than the target value or the maximum duration is reached, end the process; otherwise, return to step S1.

2. The dynamic control method according to claim 1, characterized in that: The tinting index is obtained through the following process: the acquired tooth image is converted to the LAB color space, the color difference between the tooth area and the standard whiteness is calculated, and the tinting index is obtained after normalization.

3. The dynamic control method according to claim 1, characterized in that: The temperature state is the difference between the predicted temperature and the oral baseline temperature. The predicted temperature is calculated based on a discretized model of the law of conservation of energy, according to real-time temperature, light power, photothermal conversion coefficient, and heat dissipation coefficient.

4. The dynamic control method according to claim 1, characterized in that: The comprehensive status index Calculated using the following formula: ； in, , , These are the weighting coefficients. The color index at time t; Let be the temperature state at time t, which is the difference between the predicted temperature at time t and the oral baseline temperature. Let t represent the user's comfort level at time t.

5. The dynamic control method according to claim 1, characterized in that: In step S5, the objective function is: ； Constraints: , ; in, , , As weight; This is the safety threshold.

6. The dynamic control method according to claim 5, characterized in that: The weight , , It can be updated online using gradient descent, with the update process based on a loss function that includes whitening effect error and user comfort penalty.

7. The dynamic control method according to claim 6, characterized in that: The weights are updated by dynamically balancing them in real time by calculating the gradient of the loss function. , , The expression is as follows: i=1,2,3; in, The weight value at time t+1. Let be the weight value at time t, and η be the learning rate. This is the gradient of the loss function with respect to the weights.

8. The dynamic control method according to claim 7, characterized in that: The loss function The expression is as follows: ; in, The new pigmentation index after applying whitening parameters. The target value for the coloring index; The target value for comfort; Apply a weighted penalty to comfort levels.

9. A dynamic control system for a teeth whitening device based on multimodal perception, characterized in that: include: The main control module is used to execute the control algorithm; Image sensor used to acquire images of teeth and calculate the staining index; Temperature sensor used to monitor tooth surface temperature; The user feedback module is used to receive user comfort ratings. The light source module is used to provide whitening light. The main control module is configured to: integrate the coloring index, the temperature state obtained based on the temperature prediction model, and the user comfort state to generate a comprehensive state index, and dynamically solve the objective function based on the comprehensive state index to optimize the whitening power parameters; The dynamic control system performs the dynamic control method as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the dynamic control method as described in any one of claims 1-8.

Images