Intelligent facial skin care monitoring management system

Abstract

The invention discloses an intelligent facial skin care monitoring management system which is composed of a hardware layer, a data layer, an algorithm layer and an application layer, the hardware layer collects facial skin images and external environment data through an RGB camera, a UV light sensor, a depth sensor and an environment sensor, and the data layer achieves encryption transmission and cloud storage of the collected data. The algorithm layer performs skin state detection, classification and trend prediction based on multi-modal data fusion, a deep convolutional neural network and a time sequence model, and the application layer provides skin care scheme suggestions and virtual trial functions through a personalized recommendation algorithm. The system can detect the multi-dimensional features of the skin at high precision, generate a dynamic skin care scheme, predict the change trend of the skin state, and improve the user experience through a visual interface and an interaction function. The method has the advantages of being high in detection precision, scientific in recommendation scheme, accurate in trend prediction, safe in user privacy and the like, and can be widely applied to skin care product research and development, personalized skin care management and beauty industries.

Description

Technical Field

[0001] The present invention relates to the technical field of skin monitoring, and particularly to an intelligent facial skin care monitoring and management system. Background Art

[0002] With the continuous growth of people's demand for facial skin care, the selection and use of skin care products have become an important part of daily life. However, there are the following technical deficiencies in current facial skin care management:

[0003] Limited detection accuracy: Traditional skin care solutions usually rely on manual observation, single sensors or simple hardware devices, and it is difficult to comprehensively capture multi-dimensional features of the skin, such as skin color, pores, skin texture and the distribution of skin spots, etc.

[0004] Lack of personalization: Existing skin care solutions are mostly fixed templates, and it is difficult to dynamically adjust according to individual skin conditions, external environmental conditions (such as temperature, humidity, ultraviolet intensity) and user preferences, resulting in unsatisfactory skin care effects.

[0005] Lack of dynamic analysis and trend prediction capabilities: Existing technologies are difficult to effectively record and analyze historical data of skin conditions, nor can they predict future changes in skin conditions, and cannot meet the needs of users for long-term skin care management.

[0006] Insufficient user experience: In traditional skin care management, the functions of the interaction between users and the system are relatively single, lacking an intelligent user experience based on virtual trial, social interaction and educational functions.

[0007] Based on the above problems, there is an urgent need for a new intelligent facial skin care monitoring and management system that can integrate multi-modal data collection, artificial intelligence analysis and personalized skin care plan recommendation to meet users' skin care needs for high precision, multi-function and personalization. Summary of the Invention

[0008] The purpose of the present invention is to provide an intelligent facial skin care monitoring and management system. This system collects multi-modal data through hardware, and combines advanced artificial intelligence and deep learning algorithms to provide accurate skin condition analysis and personalized skin care plan recommendation.

[0009] To achieve the above object, the present invention is implemented according to the following technical solution:

[0010] The present invention includes a hardware layer, a data layer, an algorithm layer and an application layer. The hardware layer includes a multi-sensor module for collecting facial images, detecting skin spots, ultraviolet damage, skin texture, pore size, environmental temperature and humidity. The data layer includes data storage, data transmission and a cloud database; the algorithm layer includes a skin condition classification algorithm, and the application layer includes a user interaction interface, a skin care plan recommendation algorithm and a dynamic skin change prediction algorithm.

[0011] The hardware layer includes an RGB camera, a UV light detection sensor, a depth sensor, and an environmental sensor. The RGB camera captures high-resolution facial images and analyzes skin color and texture. The UV light detection sensor detects skin spots and UV damage under the skin. The depth sensor measures the surface texture and pore size of the skin. The environmental sensor detects the impact of temperature, humidity, and PM2.5 in the external environment. After real-time data collection by the hardware layer, the collected data is preprocessed to reduce latency and dynamically adjust the exposure and lighting of the RGB camera.

[0012] The RGB camera, UV light detection sensor, depth sensor, and environmental sensor are integrated on a makeup mirror.

[0013] The beneficial effects of the present invention are as follows:

[0014] The present invention is an intelligent facial skin care monitoring and management system. Compared with the prior art, the technical effects of the present invention are as follows:

[0015] High-precision skin detection: By integrating an RGB camera, a UV light sensor, a depth sensor, and an environmental sensor, multi-modal data collection is achieved to comprehensively detect the facial skin condition, covering multi-dimensional features such as skin color, pores, skin texture, skin spots, and environmental impact.

[0016] Dynamic optimization and personalized recommendation: Based on multi-task learning algorithms and collaborative filtering recommendation models, the system can dynamically adjust the recommendation scheme according to the user's real-time skin condition, skin care goals, and preferences, and provide scientific and personalized skin care product and lifestyle suggestions for users.

[0017] Long-term trend prediction: Using a time series analysis model (such as LSTM) to learn and model the historical data of skin conditions, predict the future changes in skin conditions, help users formulate long-term skin care plans, and achieve seamless connection from short-term care to long-term management.

[0018] Enhanced user experience: Provide a user-friendly interaction interface, including visualization of skin health reports, historical data retrieval, virtual try-on, and social interaction functions, improve the user experience, and enhance user stickiness.

[0019] Security and privacy protection: Through encrypted data transmission and distributed storage, ensure the security of user personal data, and at the same time support users to manage data usage permissions, enhancing user trust.

[0020] Through the deep integration of hardware and software, the present invention runs skin condition detection, analysis, recommendation, optimization, and trend prediction throughout the entire skin care process, significantly improving the scientific nature and efficiency of skin care management. Compared with the prior art, it has obvious technological progressiveness and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the system structure principle block diagram of the present invention. Specific embodiments

[0022] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention, but do not limit the present invention.

[0023] As Figure 1 shown: The present invention includes a hardware layer, a data layer, an algorithm layer, and an application layer. The hardware layer includes a multi-sensor module that collects facial images, detects skin spots, ultraviolet damage, skin texture, pore size, and environmental temperature and humidity. The data layer includes data storage, data transmission, and a cloud database. The algorithm layer includes a skin state classification algorithm, and the application layer includes a user interface, a skin care plan recommendation algorithm, and a dynamic skin change prediction algorithm.

[0024] The hardware layer includes an RGB camera, a UV light detection sensor, a depth sensor, and an environmental sensor. The RGB camera captures high-resolution facial images and analyzes skin color and texture. The UV light detection sensor detects skin spots and ultraviolet damage under the skin. The depth sensor measures skin surface texture and pore size. The environmental sensor detects temperature, humidity, PM2.5, and other external environmental impacts. After real-time data collection by the hardware layer, the collected data is preprocessed to reduce latency and dynamically adjust the exposure and lighting of the RGB camera.

[0025] The RGB camera, UV light detection sensor, depth sensor, and environmental sensor are integrated on a makeup mirror. According to actual needs, a near-infrared light sensor can also be added to detect skin moisture content and sebum secretion. Because near-infrared light can penetrate the skin surface and obtain deeper skin information, it helps to more comprehensively evaluate the skin state.

[0026] For the RGB camera, a dual-camera design can also be adopted, one for taking frontal facial images and the other for taking side images to more accurately measure skin surface texture and pore size.

[0027] The algorithm for the data preprocessing is as follows:

[0028] I fused (x, y) = α·I RGB (x, y) + β·I UV (x, y)

[0029] Where: I RGB (x, y): The intensity value of the RGB image at pixel (x, y); I UV(x, y): The intensity value of the UV image at pixel (x, y)

[0030] Its purpose is to perform weighted fusion on the RGB image and the UV image to generate an image that comprehensively reflects the skin condition for subsequent analysis.

[0031] α, β: Weight parameters, determined by the following normalization method:

[0032]

[0033] σ RGB and σ UV are the standard deviations of the RGB and UV images respectively, used to dynamically balance the weights. By dynamically adjusting the weight α in this way, the fused image can better balance the contributions of the RGB and UV images, avoiding the influence of one image being too strong or too weak in information on the fusion effect.

[0034] The depth sensor measures the skin surface texture and pore size through a multi-scale analysis algorithm, as follows:

[0035]

[0036] where: LBP s (x, y): The LBP value of pixel (x, y) at scale s; P: The number of neighboring pixels; I s (x, y): The intensity value of pixel (x, y) at scale s; I s (x i , y i ): The intensity value of the i-th neighboring pixel at scale s; 1(·): The indicator function, which returns 1 when the condition is true and 0 otherwise;

[0037] The above formula is used to calculate the local binary pattern features. By comparing the intensity relationship between neighboring pixels and the central pixel, a binary pattern is generated to describe the local texture.

[0038] By fusing the LBP values at multiple scales, more details can be captured at different resolutions:

[0039]

[0040] where: γ s : Is the weight, assigned according to the importance of each scale. By performing weighted fusion on the LBP values at different scales, more details can be captured at different resolutions, thus more comprehensively describing the skin texture and pore size.

[0041] The skin condition classification algorithm is based on multi-task learning of deep convolutional neural networks, predicting multiple skin problems simultaneously. It uses a backbone network to extract shared features, and multi-task branches to predict each skin problem respectively. The loss function is as follows:

[0042]

[0043] Where: T: The total number of tasks, including acne detection and freckle detection; L t : The loss function of task t; λ t : The task weight, used to balance the importance of different tasks.

[0044] The above formula is used for multi-task learning. Shared features are extracted through the backbone network, and then each skin problem is predicted through multi-task branches.

[0045] Dynamic adjustment of task weights:

[0046]

[0047] Var(L t ) is the variance of the task loss, used to balance training. By dynamically adjusting the task weights, the training of the model on different tasks becomes more balanced, avoiding the overall training effect being affected by an overly large or small loss in a certain task.

[0048] The skin care plan recommendation algorithm recommends suitable skin care products according to user characteristics and skin problems. The formula for predicting the skin care product score:

[0049]

[0050] Where: The predicted score of user u for skin care product i; w 0 : The global bias term; w j : The weight of feature j; x u，i，j : The value of user u and skin care product i on feature j; v j 、v k : The embedding vectors of features j and k, and the inner product <v j ,v k > is used to model the interaction.

[0051] By modeling the interaction between user features and skin care product features, the satisfaction of users with skin care products is predicted, thus realizing personalized recommendation.

[0052] The dynamic skin change prediction algorithm predicts future changes in skin conditions and provides long-term skin care suggestions. It is based on the skin trend prediction of a time series model. The core state update formula for the skin trend prediction of the time series model:

[0053] h t =f(Wx ·x t +W h ·h t-1 +b)

[0054] Where: h t : The hidden state at the current time step, x t : The current input, including skin state metrics and environmental data; W x , W h : The weight matrix; b: The bias term; f(·): The activation function, used to introduce non-linearity.

[0055] The above formula is the core state update formula of a time series model (such as LSTM), used to update the current state based on the current input and the state at the previous moment.

[0056] The predicted skin trend is:

[0057] y t =W y ·h t +b y

[0058] Where: y t : The predicted skin state, including sebum secretion amount and age spot area; W y , b y : The weights and bias of the output layer. Used to map the hidden state h t to the final prediction result, thereby realizing the prediction of the future skin state.

[0059] Example: Skin state detection and analysis

[0060] The user performs a facial scan through a smart makeup mirror or a portable device.

[0061] Hardware layer: The RGB camera captures high-resolution skin images and analyzes skin color uniformity. The UV light sensor detects age spots and UV damage under the skin. The depth sensor captures the skin surface texture and pore size. The environmental sensor records temperature, humidity, and air quality parameters.

[0062] Data layer: The sensor data is uploaded to the cloud database for storage after local processing. The data is transmitted encrypted to ensure user privacy. A hybrid encryption algorithm is used, combining the advantages of symmetric encryption and asymmetric encryption, to improve the security of data transmission. For example, the asymmetric encryption algorithm (such as RSA) is used to encrypt the symmetric encryption key, and then the symmetric encryption algorithm (such as AES) is used to encrypt the data, which can not only ensure the encryption speed but also ensure the security of the data.

[0063] In terms of cloud storage, in addition to data encryption, distributed storage technology can also be adopted to disperse data storage across multiple servers and set up access permissions and data backup mechanisms to prevent data loss and unauthorized access.

[0064] In the data preprocessing stage, a denoising step can also be added. The adaptive filtering algorithm is used to remove the noise data collected by the sensor to improve the data quality. For example, for RGB images, the bilateral filtering algorithm can be used to remove noise while retaining the edge information of the image; for UV images and depth images, the median filtering algorithm can be used to remove noise.

[0065] Regarding the normalization method of weight parameters, machine learning algorithms can be considered to automatically adjust the weight parameters according to historical data to better balance the contributions of RGB images and UV images and improve the accuracy of skin condition detection.

[0066] Algorithm layer: Use the multi-modal data fusion algorithm to synthesize RGB and UV images into a high-precision skin condition map:

[0067] I fused (x, y) = α·I RGB (x, y) + β·I UV (x, y)

[0068] Use a deep learning model (such as SkinNet based on ResNet) to classify skin problems (skin spots, acne):

[0069] P problem = softmax(W·f(I fused ) + b)

[0070] Output the skin health score and the problem classification result.

[0071] Optimization of the skin condition classification algorithm:

[0072] In the backbone network, an improved ResNet architecture is adopted to increase the number of residual connections, improve the depth and complexity of the network, and thus better extract the features of skin images. At the same time, in the multi-task branch, different activation functions and loss functions are designed for each task and optimized according to the characteristics of the task. For example, for the acne detection task, the ReLU activation function and binary cross-entropy loss function can be used; for the skin spot detection task, the LeakyReLU activation function and mean squared error loss function can be used.

[0073] Introduce an attention mechanism to enable the network to automatically focus on key areas in skin images, such as acne, skin pigmentation, etc., improving the accuracy and robustness of detection. For example, a combination of channel attention and spatial attention mechanisms can be used to weight the feature maps of skin images, highlighting important features.

[0074] Personalized skin care plan recommendation:

[0075] The user inputs skin care goals (such as reducing skin pigmentation) and preferences (such as suitable for sensitive skin). The system recommends skin care products and lifestyle adjustment plans based on the detection results and the user's historical data:

[0076] Skin care product rating prediction formula:

[0077]

[0078] Users can virtually try out recommended products through the augmented reality (AR) function.

[0079] Optimization of skin care plan recommendation algorithm:

[0080] In the determination of features j and k, automatic feature extraction methods, such as autoencoders or variational autoencoders, can be used to automatically learn features from the user's historical data and skin condition data, avoiding the subjectivity and limitations of manually designed features.

[0081] Introduce a user feedback mechanism. According to the user's feedback on the usage effects of recommended skin care products, dynamically adjust the parameters of the recommendation algorithm to improve the accuracy and personalization of the recommendation. For example, if the user feedbacks that a certain skin care product has poor effects, reduce the rating weight of this skin care product, and at the same time adjust the similarity calculation method between user features and skin care product features to better match the user's actual needs.

[0082] Dynamic skin change prediction:

[0083] The system conducts long-term skin health tracking on users and records historical data.

[0084] Use a time series prediction model (LSTM) to analyze the changing trends of skin indicators:

[0085] h t =f(W x ·x t +W h ·h t-1 +b)

[0086] Output the predicted value of the future skin state:

[0087] y t = W y ·h t + b y

[0088] Dynamically adjust the skin care plan according to the prediction results. For example, increase the moisturizing plan in autumn and winter seasons.

[0089] Optimization of the dynamic skin change prediction algorithm:

[0090] In terms of the training data of the time series model, in addition to using the historical data of the user's own skin condition, similar data of other users can also be introduced as auxiliary training data. Through the method of transfer learning, the generalization ability and prediction accuracy of the model can be improved. For example, the data of users with similar skin types and living environments can be clustered, and then the clustered data can be used as auxiliary training data to train the time series model together with the user's own data.

[0091] Add an uncertainty estimation module to quantify the uncertainty of the prediction results and provide more reliable prediction information for users. For example, the Bayesian Neural Network or Monte Carlo Dropout method can be used to estimate the uncertainty of the prediction results of the time series model. When the prediction uncertainty is high, remind the user to refer to the prediction results carefully and recommend that the user consult a professional dermatologist in time.

[0092] Application layer:

[0093] User interface: In terms of the visualization of the skin health report, interactive charts such as line charts, bar charts, and pie charts are adopted. Users can view detailed data and trend analysis through operations such as hovering and clicking with the mouse. For example, in a line chart, users can view the change trend of the skin health score over time. When the mouse hovers over a certain time point, the specific skin condition data at that time point, such as the area of skin spots and the amount of sebum secretion, can be displayed.

[0094] Add a skin health goal setting function. Users can set short-term and long-term skin health goals according to their own needs, such as reducing the area of skin spots and increasing the skin moisture content. The system automatically generates a personalized skin care plan according to the goals set by the user, and real-time tracks the changes in the user's skin condition to dynamically adjust the skin care plan to help users better achieve their skin health goals.

[0095] Virtual trial function optimization: The virtual trial function is implemented using augmented reality (AR) technology. The camera captures the user's facial image in real time, and the recommended skincare product effects are rendered onto the user's face in real time, allowing the user to intuitively see the usage effects. For example, for a whitening skincare product, the whitened skin tone effect can be simulated on the user's face in real time; for an anti-wrinkle skincare product, the effect of reduced wrinkles can be simulated.

[0096] Add a virtual trial effect comparison function. Users can compare the virtual trial effects with the actual usage effects, and the system automatically adjusts the parameters of the virtual trial model according to the user's feedback to improve the accuracy and reliability of the virtual trial effects.

[0097] Example 2:

[0098] The user is a 30-year-old female with combination skin. The main skin problems are pigmentation and enlarged pores. The user hopes to reduce the area of pigmentation, shrink pores, and maintain the water-oil balance of the skin.

[0099] Skin condition detection:

[0100] The user uses a smart makeup mirror to perform a facial scan every morning. The RGB camera captures high-resolution skin images to analyze skin color uniformity and skin texture; the UV light detection sensor detects pigmentation and UV damage in the lower layer of the skin; the depth sensor captures the skin surface texture and pore size; the environmental sensor records temperature, humidity, and air quality parameters.

[0101] The system preprocesses the collected data to remove noise and synthesizes a high-precision skin condition map through a multi-modal data fusion algorithm. For example, using the above data preprocessing algorithm and the multi-scale analysis algorithm of the depth sensor, a skin condition map is obtained, showing that the user's pigmentation area is 10 square millimeters, the average pore diameter is 0.2 millimeters, and the skin color uniformity score is 70 points (out of 100).

[0102] Skin condition classification and analysis:

[0103] The system uses a multi-task learning algorithm based on a deep convolutional neural network to classify skin problems. The backbone network extracts shared features, and the multi-task branches respectively predict pigmentation and pore problems. For example, through the trained SkinNet model, the output of the user's pigmentation detection result is moderate pigmentation, and the pore detection result is enlarged pores.

[0104] The system calculates the skin health score based on the classification results, comprehensively considering factors such as pigmentation, pores, and skin color uniformity, and obtains the user's current skin health score of 65 points (out of 100).

[0105] Personalized skincare plan recommendation:

[0106] The user inputs skin care goals of reducing age spots and shrinking pores in the system, with a preference for sensitive skin adaptation. Based on the user's skin condition detection results, skin care goals and preferences, combined with the user's historical data (such as the skin condition change trend in the past month), the system recommends suitable skin care products and lifestyle adjustment plans.

[0107] In terms of skin care product recommendations, the system selects products with higher scores from the skin care product database according to the predicted skin care product scoring formula. For example, a whitening essence is recommended, and the predicted score for the user's evaluation of this product is 85 points; a pore-tightening toner is recommended, and the predicted score is 80 points. At the same time, the system also recommends that the user clean the face with a mild cleansing product every morning and evening, and avoid using strongly irritating cosmetics.

[0108] In terms of lifestyle adjustment plans, the system advises the user to ensure sufficient sleep every day and avoid staying up late; drink plenty of water to keep the skin sufficiently hydrated; perform a deep cleansing mask treatment once a week to help remove dirt and excess oil on the skin surface.

[0109] The user can virtually try out the recommended whitening essence and toner through the augmented reality (AR) function. The system renders the virtual trial effect onto the user's face in real time, allowing the user to intuitively see the skin improvement effect after use. For example, after the virtual trial, the user can see that the area of age spots has decreased to 8 square millimeters, the average pore diameter has shrunk to 0.18 millimeters, and the skin tone uniformity score has increased to 75 points.

[0110] Dynamic skin change prediction and plan adjustment:

[0111] The system conducts long-term skin health tracking on the user and records the daily skin condition data. A time series prediction model (LSTM) is used to analyze the change trend of skin indicators. For example, based on the skin condition data in the past month, it is predicted that in the next month, the area of the user's age spots will decrease to 7 square millimeters, the average pore diameter will shrink to 0.17 millimeters, and the skin tone uniformity score will increase to 80 points.

[0112] According to the prediction results, the system dynamically adjusts the skin care plan. For example, if it is predicted that the skin moisture content of the user may decrease in autumn and winter, the system will recommend in advance that the user increase the usage frequency of moisturizing skin care products and recommend a moisturizing cream suitable for autumn and winter, with a predicted score of 90 points. At the same time, the system will also remind the user to pay attention to adjusting the indoor air humidity to avoid other problems caused by dry skin.

[0113] User feedback and plan optimization:

[0114] After using the recommended skin care products for a period of time, the user feedbacks the usage effects through the system. For example, the user feedbacks that after using the whitening essence, the area of skin pigmentation has significantly decreased, but after using the toner, there is a slight tingling sensation on the skin. Based on the user's feedback, the system adjusts the skin care product recommendation plan, reduces the scoring weight of the toner, and at the same time screens out other pore-tightening products suitable for sensitive skin from the database for recommendation, and re-adjusts the lifestyle adjustment plan, suggesting that the user conduct a local skin test before using the toner to ensure that the product suits their skin.

[0115] The system records the user's feedback data for optimizing the parameters of the recommendation algorithm and prediction model, and continuously improves the accuracy and personalization of the recommendation. For example, by analyzing a large amount of user feedback data, it is found that a certain brand of toner has a greater irritation to sensitive skin. The system will automatically adjust the scoring weight of this product, reduce its ranking in the recommendation list, and at the same time optimize the feature extraction and similarity calculation methods to better match the actual needs of users.

[0116] Through the above embodiments, the working process and effects of the intelligent facial skin care monitoring and management system of the present invention in practical applications are demonstrated, reflecting its advantages such as high-precision skin detection, dynamic optimization and personalized recommendation, long-term trend prediction, and enhanced user experience, and can meet the needs of users for personalized skin care management.

[0117] The technical solution of the present invention is not limited to the restrictions of the above specific embodiments. Any technical deformation made according to the technical solution of the present invention falls within the protection scope of the present invention.

Claims

1. An intelligent facial skin care monitoring and management system, characterized by: It includes a hardware layer, a data layer, an algorithm layer and an application layer. The hardware layer includes a multi-sensor module to collect facial images, detect spots, ultraviolet damage, skin texture, pore size, and ambient temperature and humidity. The data layer includes data storage, data transmission and a cloud database. The algorithm layer includes a skin condition classification algorithm, and the application layer includes a user interaction interface, a skin care plan recommendation algorithm and a dynamic skin change prediction algorithm.

2. The intelligent facial skin care monitoring and management system according to claim 1 is characterized by: The hardware layer includes an RGB camera, a UV light detection sensor, a depth sensor, and an environmental sensor. The RGB camera captures high-resolution facial images and analyzes skin color and skin texture. The UV light detection sensor detects spots and UV damage under the skin. The depth sensor measures skin surface texture and pore size. The environmental sensor detects temperature, humidity, and PM2.5 external environmental influences. After real-time data collection, the hardware layer pre-processes the collected data to reduce latency and dynamically adjust the exposure and lighting of the RGB camera.

3. The intelligent facial skin care monitoring and management system according to claim 2 is characterized by: The RGB camera, UV light detection sensor, depth sensor, and environmental sensor are integrated on the makeup mirror.

4. The intelligent facial skin care monitoring and management system according to claim 2 is characterized by: The data preprocessing algorithm is as follows: I fused (x,y)=α·I RGB (x,y)+β·I UV (x,y) Where: I RGB (x, y): intensity value of the RGB image at pixel (x, y); I UV (x, y): intensity value of UV image at pixel (x, y), α, β: weight parameters, determined by the following normalization method: σ RGB and σ UV They are the standard deviations of RGB and UV images, respectively, used for dynamic balancing weights.

5. The intelligent facial skin care monitoring and management system according to claim 2 is characterized by: The depth sensor measures the skin surface texture and pore size through a multi-scale analysis algorithm as follows: Among them: LBP s (x, y): LBP value of pixel (x, y) when scale is s; P: number of neighborhood pixels; I s (x, y): intensity value of pixel (x, y) at scale s; I s (x i ,y i ): the intensity value of the i-th neighborhood pixel at scale s; 1(·): indicator function, returns 1 if the condition is true, otherwise returns 0; By fusing LBP values ​​at multiple scales, more details can be captured at different resolutions: Where: γ s : is the weight, which is allocated according to the importance of each scale.

6. The intelligent facial skin care monitoring and management system according to claim 1 is characterized by: The skin condition classification algorithm is based on multi-task learning of deep convolutional neural networks, and predicts multiple skin problems at the same time. The backbone network is used to extract shared features, and the multi-task branches predict each skin problem separately. The loss function is as follows: Where: T: total number of tasks, including acne detection and spot detection; L t : loss function of task t; λ t : Task weight, dynamic adjustment: Var(L t ) is the variance of the task loss, which is used to balance the training.

7. The intelligent facial skin care monitoring and management system according to claim 1 is characterized by: The skin care solution recommendation algorithm recommends suitable skin care products based on user characteristics and skin problems, and the formula for predicting the skin care product score is: in: User u's rating prediction for skin care product i; w0: global bias term; w j : The weight of feature j; x u，i，j : The value of user u and skin care product i on feature j; v j 、v k : Embedding vector of features j and k, inner product <v j , v k >For modeling interactions.

8. The intelligent facial skin care monitoring and management system according to claim 1 is characterized by: The dynamic skin change prediction algorithm predicts future changes in skin conditions, provides long-term skin care recommendations, and predicts skin trends based on a time series model. The core state update formula for the skin trend prediction of the time series model is: h t =f(W x ·x t +W h ·h t-1 +b) Where: h t : The hidden state of the current time step, x t : Current input, including skin condition indicators and environmental data; W x , W h : weight matrix; b: bias term; f(·): activation function; the output skin trend prediction is: y t =W y ·h t +b y Where: y t : Predicted skin condition, including oil secretion and pigmentation area; W y 、b u : Output layer weights and biases.

Images