A makeup personalization recommendation method and system based on skin texture detection

Abstract

The application is suitable for the field of image recognition, and discloses a makeup personalization recommendation method and system based on skin quality detection. The method extracts multi-dimensional skin physiological parameters by multi-spectral partition sampling to construct microscopic state features, solving the tolerance threshold evaluation misalignment problem caused by single data dimension in traditional methods. The basic fragility index is obtained by analyzing the thickness of the stratum corneum and the fragility of the microvessels, and the personalized skin tolerance threshold interval is quantified by calculating the deviation of the red blood filament density and the epidermal pH value. Through the analysis and quantification of the candidate product ingredient text, the irritation value is calculated and compared with the tolerance threshold, realizing the conflict judgment from the microscopic physiological level to the chemical component level, avoiding the sensitization risk at the source, and significantly improving the safety of the recommendation.

Description

Technical Field

[0001] This application belongs to the field of image recognition, and in particular relates to a personalized beauty recommendation method and system based on skin texture detection. Background Technology

[0002] Personalized beauty recommendations are intelligent recommendation services that match cosmetics, skincare, and makeup products to suit an individual's skin type, needs, and preferences. They are widely used in beauty apps, e-commerce platforms, smart skin analysis at department store counters, and beauty mini-programs.

[0003] Traditional beauty recommendation technologies rely primarily on user-submitted skin type labels or single-dimensional image analysis for product matching. They cannot assess the skin's true tolerance threshold to specific chemical ingredients, making it highly susceptible to recommending products containing sensitizing ingredients to users with sensitive skin, thus posing serious skin health risks.

[0004] The information disclosed in the background section is only for enhancing the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] In view of this, this application provides a method and system for personalized beauty recommendations based on skin type detection, which can solve the problem in related technologies that products containing allergenic ingredients are easily recommended to users with sensitive skin, causing serious skin health risks.

[0006] The first aspect of this application provides a personalized beauty recommendation method based on skin type analysis, the method comprising:

[0007] Obtain the user's facial skin image;

[0008] Multispectral partitioning sampling is performed on the facial skin image to extract multidimensional skin physiological parameters as the user's micro-state features. The micro-state features include stratum corneum thickness, sebum secretion rate, red blood vessel density, pore size parameters, melanin deposition, water content index, microvascular fragility, and epidermal pH value.

[0009] Based on the microscopic characteristics, the user's skin tolerance threshold range is determined, and the ingredient text data of candidate beauty products is obtained.

[0010] The ingredient text data is parsed and quantified to separate chemical component characteristics and concentration characteristics. Based on the chemical component characteristics and concentration characteristics, the irritation value of the candidate cosmetic product is calculated to obtain the sensitization risk characteristics.

[0011] If the sensitization risk characteristics exceed the mapping threshold of the skin tolerance threshold range, and the sensitization markers in the chemical composition characteristics overlap with the user's physiological characteristics, then a micro-conflict is determined, and products with micro-conflicts are eliminated to obtain the remaining products.

[0012] The remaining products are filtered to obtain a set of safe beauty products, and the degree of fit between the target products in the set of safe beauty products and the microscopic features is calculated.

[0013] The results are sorted according to the degree of fit to obtain the recommendation results.

[0014] Optionally, in a first implementation of the first aspect of this application, determining the user's skin tolerance threshold range based on the microscopic state characteristics includes:

[0015] The pre-trained association mapping model is invoked, and the stratum corneum thickness and the microvascular fragility are used as the first input group to calculate the nonlinear correlation between the two and obtain the basic fragility index.

[0016] Using the density of red blood vessels and the pH value of the epidermis as the second input group, the absolute value of the difference between the two is calculated to obtain the environmental stimulus offset.

[0017] Using the sebum secretion rate and the pore size parameter as the third input group, regression analysis was used to determine the erosion coefficient of oil accumulation on the skin barrier.

[0018] Using the melanin deposition amount and the water content index as the fourth input group, the attenuation factor of pigment oxidative stress response to tolerance threshold is calculated.

[0019] The basic vulnerability index, the environmental stimulus offset, the erosion coefficient, and the attenuation factor are weighted and fused to obtain a comprehensive tolerance index.

[0020] The user's skin tolerance threshold range is obtained by indexing and scaling the comprehensive tolerance index in a preset threshold range table.

[0021] Optionally, in the second implementation of the first aspect of this application, the step of parsing and quantifying the component text data to separate chemical component features and concentration features includes:

[0022] Sequence labeling and entity extraction are performed on the component text data to obtain the component entity sequence. The numerical information in the component text data is parsed to extract the concentration percentage and permeation rate coefficient of each component, and a concentration feature vector is obtained.

[0023] The component entity sequence is linked with a preset chemical substance knowledge graph to extract the chemical properties and risk tags of each component, thereby obtaining a chemical component feature vector. The chemical component feature vector is then integrated with the concentration feature vector to obtain the chemical component features and concentration features of the candidate cosmetic product.

[0024] Optionally, in a third implementation of the first aspect of this application, the step of calculating the irritation value of the candidate cosmetic product based on the chemical composition characteristics and the concentration characteristics to obtain the sensitization risk characteristics includes:

[0025] Extract the alcohol content ratio and surfactant strength from the characteristics of the chemical components;

[0026] The cumulative risk value is obtained by calculating the weighted sum of the alcohol content ratio and the surfactant strength, and summing the preservative residue and heavy metal leaching in the concentration characteristics. The weighted sum is calculated as follows: alcohol content ratio × alcohol corresponding weight + surfactant strength × surface activity corresponding weight. When the alcohol content ratio is greater than a preset ratio, the alcohol corresponding weight is taken as the first weight coefficient; otherwise, the alcohol corresponding weight is taken as the second weight coefficient. When the surfactant strength is greater than a preset strength, the surface activity corresponding weight is taken as the third weight coefficient; otherwise, the surface activity corresponding weight is taken as the fourth weight coefficient.

[0027] The stimulus value is determined based on the superposition weight and the cumulative risk value;

[0028] Based on the mapping relationship between the irritation value and the preset risk level table, the sensitization risk characteristics of the candidate cosmetic products are determined. The risk level table is divided into low risk, medium risk, high risk, and very high risk in ascending order of irritation value. The low risk, medium risk, high risk, and very high risk correspond to differentiated risk warning strategies.

[0029] Optionally, in the fourth implementation of the first aspect of this application, calculating the numerical value of the degree of fit between the target product in the set of safe beauty products and the microscopic state characteristics includes:

[0030] The Euclidean distance between the permeation rate coefficient of the target product and the moisture content index is calculated to obtain the adsorption compatibility. The product between the number of fragrance additives in the target product and the amount of melanin deposition is calculated to obtain the reaction intensity.

[0031] By combining the adsorption adaptability and the reaction intensity, the multimodal feature vector of the target product is obtained;

[0032] Calculate the cosine similarity between the multimodal feature vector and the microstate feature vector to obtain a numerical value of the degree of fit.

[0033] Optionally, in the fifth implementation of the first aspect of this application, the filtering of the remaining products includes:

[0034] Extract the pH value of the remaining product and calculate the absolute deviation between the pH value and the pH value of the skin to obtain the pH deviation amount;

[0035] Determine whether the acid-base deviation is greater than a preset acid-base tolerance threshold, wherein the acid-base tolerance threshold is obtained by dynamically correcting the base threshold and the epidermal pH value;

[0036] If the acid-base deviation exceeds the acid-base tolerance threshold, the remaining products are deemed unqualified and discarded.

[0037] If the acid-base deviation is not greater than the acid-base tolerance threshold, the remaining products are deemed qualified and included in the safe beauty product set.

[0038] Optionally, in the sixth implementation of the first aspect of this application, the filtering of the remaining products to obtain a set of safe beauty products includes:

[0039] Based on the density of red blood vessels and the fragility of microvessels, assess the user's skin sensitivity level and obtain the formula mildness rating and allergen exclusion label of the remaining products;

[0040] If the skin sensitivity level is Level 1, then only products with the highest formula mildness rating and labeled with the all-allergen exclusion mark will be retained to obtain the safe beauty product set.

[0041] If the skin sensitivity level is level 2, products with a formula mildness rating lower than the preset sensitivity threshold or without the all-allergen exclusion label are excluded to obtain the set of safe beauty products.

[0042] If the skin sensitivity level is "not sensitive", then the remaining products are retained as part of the safe beauty product set.

[0043] Optionally, in the seventh implementation of the first aspect of this application, the step of calculating the product between the number of fragrance types added to the target product and the amount of melanin deposition to obtain the reaction intensity includes:

[0044] If the amount of melanin deposited is greater than a preset deposition threshold, a first weighting coefficient is assigned to the number of fragrance additives.

[0045] If the amount of melanin deposition is not greater than the deposition threshold, then a second weighting coefficient is assigned to the number of fragrance additives, wherein the second weighting coefficient is less than the first weighting coefficient;

[0046] The reaction intensity is obtained by weighting the number of flavoring types added according to the assigned weighting coefficients.

[0047] Optionally, in the eighth implementation of the first aspect of this application, the step of sorting according to the matching degree value to obtain the recommendation result includes:

[0048] The initial sorting result is obtained by sorting the results according to the compatibility scores.

[0049] Extract a preset number of candidate products that rank highly in the initial sorting list, and obtain preset applicable parameters for the candidate products, including oil control efficacy level and pore minimization level.

[0050] Based on the sebum secretion rate, the pore size parameter, and the preset applicable parameters, a corresponding matching level is obtained.

[0051] Secondly, embodiments of this application provide a personalized beauty recommendation system based on skin texture detection, the system comprising:

[0052] The acquisition module is used to acquire the user's facial skin image;

[0053] The sampling module is used to perform multispectral partitioning sampling on the facial skin image and extract multi-dimensional skin physiological parameters as the user's micro-state features. The micro-state features include stratum corneum thickness, sebum secretion rate, red blood vessel density, pore size parameters, melanin deposition amount, water content index, microvascular fragility and epidermal pH value.

[0054] The determination module is used to determine the user's skin tolerance threshold range based on the microscopic state characteristics, and to obtain the ingredient text data of candidate beauty products, wherein the skin tolerance threshold range includes a basic fragility index and an environmental stimulus offset.

[0055] The calculation module is used to parse and quantify the ingredient text data, separate the chemical component characteristics and concentration characteristics, and calculate the irritation value of the candidate beauty product based on the chemical component characteristics and the concentration characteristics to obtain the sensitization risk characteristics.

[0056] The elimination module is used to determine that there is a micro-conflict if the sensitization risk characteristics exceed the mapping threshold of the skin tolerance threshold range, and the sensitization marker in the chemical component characteristics overlaps with the user's physiological characteristics, and to eliminate the products with micro-conflicts to obtain the remaining products.

[0057] A filtering module is used to filter the remaining products to obtain a set of safe beauty products, and to calculate the degree of fit between the target products in the set of safe beauty products and the microscopic features.

[0058] The recommendation module is used to sort the results according to the matching degree values ​​to obtain recommendation results.

[0059] The beneficial effects of this application's embodiments compared to traditional technologies are as follows: By extracting multi-dimensional skin physiological parameters through multispectral partitioning sampling to construct microscopic state characteristics, it solves the problem of inaccurate tolerance threshold assessment caused by the single data dimension in traditional methods. A basic fragility index is obtained by analyzing stratum corneum thickness and microvascular fragility, and a personalized skin tolerance threshold range is comprehensively quantified based on the deviation between redness density and epidermal pH value. By parsing and quantifying the ingredient text of candidate products, irritation values ​​are calculated and compared with tolerance thresholds, achieving conflict determination from the microphysiological level to the chemical composition level, avoiding sensitization risks at the source, and significantly improving the safety of recommendations. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram of an embodiment of the personalized beauty recommendation method based on skin texture detection in this application.

[0062] Figure 2 This is a schematic diagram of Embodiment 2 of the personalized beauty recommendation method based on skin texture detection in this application.

[0063] Figure 3 This is a schematic diagram of Embodiment 3 of the personalized beauty recommendation method based on skin texture detection in this application.

[0064] Figure 4 This is a schematic diagram of Embodiment 4 of the personalized beauty recommendation method based on skin texture detection in this application.

[0065] Figure 5 This is a schematic diagram of Embodiment 5 of the personalized beauty recommendation method based on skin texture detection in this application.

[0066] Figure 6 This is a schematic diagram of Embodiment Six of the Personalized Makeup Recommendation Method Based on Skin Texture Detection in this application.

[0067] Figure 7 This is a schematic diagram of Embodiment Seven of the Personalized Makeup Recommendation Method Based on Skin Texture Detection in this application.

[0068] Figure 8 This is a schematic diagram of Embodiment 8 of the personalized beauty recommendation method based on skin texture detection in this application.

[0069] Figure 9 This is a schematic diagram of Embodiment Nine of the Personalized Makeup Recommendation Method Based on Skin Texture Detection in this application.

[0070] Figure 10 This is a schematic diagram of one embodiment of the personalized beauty recommendation system based on skin texture detection in this application. Detailed Implementation

[0071] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be integrated in any suitable manner into one or more embodiments. In the following description, numerous specific details are set forth to give a full understanding of embodiments of this disclosure.

[0072] The terms “a,” “one,” and “” are used to indicate the existence of one or more elements / components / etc.; the terms “including” and “having” are used to indicate an open-ended meaning of inclusion and that other elements / components / etc. may exist in addition to those listed. The terms “first” and “second” are used only as markers and are not a limitation on the number of objects.

[0073] Personalized beauty recommendations are intelligent recommendation services that match cosmetics, skincare, and makeup products to suit an individual's skin type, needs, and preferences. They are widely used in beauty apps, e-commerce platforms, smart skin analysis at department store counters, and beauty mini-programs.

[0074] Traditional beauty recommendation technologies primarily rely on user-submitted skin type labels or limited questionnaire data for product matching, separating skin type analysis, ingredient screening, and virtual try-on into independent steps. This lack of objective analysis of users' real facial images makes it impossible to make coherent decisions based on a unified skin type feature dataset, thus compromising the safety of ingredients in beauty recommendations.

[0075] In view of this, this application provides a personalized beauty recommendation method and system based on skin texture detection. It constructs microscopic state characteristics by extracting multi-dimensional skin physiological parameters through multispectral partitioning sampling, solving the problem of inaccurate tolerance threshold assessment caused by the single data dimension in traditional methods. A basic fragility index is obtained by analyzing stratum corneum thickness and microvascular fragility, and a personalized skin tolerance threshold range is comprehensively quantified based on the deviation between redness density and epidermal pH value. By parsing and quantifying the text of candidate product ingredients, irritation values ​​are calculated and compared with tolerance thresholds, achieving conflict determination from the microscopic physiological level to the chemical composition level, avoiding sensitization risks at the source, and significantly improving the safety of recommendations.

[0076] To illustrate the technical solution of this application, specific embodiments are described below.

[0077] Example 1:

[0078] This disclosure provides a personalized beauty recommendation method based on skin texture analysis, such as... Figure 1 The illustrated method is a personalized beauty recommendation system based on skin texture analysis. This method may include the following steps:

[0079] Step S101: Obtain the user's facial skin image.

[0080] In the embodiments of this application, the skin texture detection-based personalized beauty recommendation system activates the image acquisition interface, calls the high-resolution camera or multispectral imaging device of the terminal to acquire the frontal facial image of the target user, and converts the image into a preset color space and resolution before storing it in the system's temporary cache area.

[0081] Optionally, a fill light can be turned on in low-light environments, or the user can be guided to complete facial posture calibration through the screen light source, so that the captured facial skin image covers the entire area including the forehead, cheeks, nose, and chin.

[0082] Step S102: Perform multispectral partition sampling on the facial skin image and extract multidimensional skin physiological parameters as the user's micro-state features. Among them, the micro-state features include stratum corneum thickness, sebum secretion rate, red blood vessel density, pore size parameters, melanin deposition amount, water content index, microvascular fragility and epidermal pH value.

[0083] In the embodiments of this application, the facial skin image is divided into six independent sampling areas according to anatomical features: forehead, left cheek, right cheek, nose, jaw, and periorbital area. Specifically, a facial key point detection algorithm is used to locate core facial anatomical landmarks, such as the glabella, the outer corners of the eyes, the corners of the mouth, the tip of the nose, and the angles of the jaw, and a normalized facial coordinate system is generated using these anatomical landmarks.

[0084] Based on predefined geometric proportion parameters, the boundaries of each region are clipped. The forehead region is defined as the area above the line connecting the eyebrows and the temporal region on both sides; the left and right cheek regions are defined as the symmetrical areas between the line connecting the eyebrows and the horizontal line of the nasal ala, and each is bounded by the vertical midline of the face; the nose region is defined as the area of ​​the bridge of the nose and the nasal ala; the jaw region is defined as the area below the horizontal line of the nasal ala to the bottom of the chin, and includes the contour area of ​​the mandible on both sides; the periorbital region is defined as the upper and lower eyelids and the periorbital transition area within a preset radius centered on the outer corners of the eyes.

[0085] Through the above coordinate mapping and region clipping, the six independent sampling areas of the forehead, left cheek, right cheek, nose, jaw, and eye area can be divided.

[0086] Pixel-level matrix sampling is performed on each sampling area in eight preset bands from 365nm to 1050nm to generate a multi-channel spectrum.

[0087] The photon penetration delay is calculated using a backscattering optical path difference algorithm based on the 850nm near-infrared band, thus determining the stratum corneum thickness. Specifically, by analyzing the time difference characteristics of photoplethysmography (PPG) signals propagating in tissues and examining the phase shift or time delay of backscattered light in the 850nm band, the average propagation path length of photons in tissues is calculated, thereby quantitatively characterizing the physical thickness of the stratum corneum.

[0088] Porphyrin fluorescence was excited using 365nm ultraviolet light, and the sebum secretion rate was calculated by quantifying the dynamic change rate of the fluorescent pixels. Specifically, by detecting the characteristic fluorescence intensity produced by porphyrin secreted by sebaceous glands under 365nm excitation light, and combining it with a skin surface oil oxidation kinetic model, the decay rate or accumulation rate of the fluorescence signal was converted into the amount of sebum secreted per unit time.

[0089] Microvascular networks were extracted using an adaptive thresholding algorithm under 560nm green light band to obtain the density of red blood vessels. Specifically, the absorption difference of green light in hemoglobin and surrounding tissues was used to binarize the image using an adaptive thresholding algorithm to segment the area ratio of the capillary network, which was then converted into the density of red blood vessels per unit area.

[0090] Edge enhancement is performed on high-frequency spatial domain images using the Laplacian operator, and concave regions are identified to derive pore size parameters. Specifically, the Laplacian operator is used to highlight high-frequency details in the image in the frequency domain, enhance the grayscale difference between the pore edges and the surrounding flat skin, extract pore contours through morphological analysis, and calculate the area and perimeter of the pores to characterize their size.

[0091] By combining the diffuse reflectance model of the 650nm red light band with the inverse calculation of the melanin absorption coefficient, the amount of melanin deposition can be obtained. Specifically, a Monte Carlo light transmission model of skin tissue at a wavelength of 650nm is established. By fitting the actual measured diffuse reflectance data, the absorption coefficient of melanin in this wavelength band is solved in reverse, which serves as a quantitative indicator of melanin concentration.

[0092] The absorbance was calculated using Beer-Lambert's law in the 970nm water characteristic absorption band to obtain the water content index. Specifically, by measuring the light intensity attenuation after light passes through skin tissue at the characteristic strong absorption peak of water molecules at 970nm, the absorbance was converted into the water molecule concentration within the tissue, thereby obtaining the local skin water content.

[0093] By combining the density of red blood vessels with the amplitude variation coefficient of the photoplethysmography (PPG) pulse wave signal, the microvascular fragility is assessed using a support vector machine (SVM) model. Specifically, the geometric distribution features of red blood vessels and the amplitude fluctuation coefficient of the PPG signal during systole and diastole are extracted as input feature vectors. A SVM classification and regression model trained on a large amount of clinical data is then used to output a microvascular rupture risk score under blood pressure fluctuations, i.e., microvascular fragility.

[0094] By analyzing the reflectance ratio of the 450nm and 500nm bands to map the skin's pH gradient, the skin's pH value can be obtained.

[0095] After normalizing the parameters of the eight dimensions to their maximum and minimum values, they are concatenated into a one-dimensional tensor to obtain the microscopic state features.

[0096] Step S103: Based on the microscopic characteristics, determine the user's skin tolerance threshold range and obtain the ingredient text data of the candidate beauty products.

[0097] In the embodiments of this application, a pre-trained association mapping model is invoked, and the stratum corneum thickness and microvascular fragility in the micro-state features are used as inputs to calculate the nonlinear correlation between the two to obtain the basic fragility index; at the same time, the red blood vessel density and epidermal pH value are used as inputs to calculate the absolute value of the difference between the two to obtain the environmental stimulus offset; the basic fragility index and the environmental stimulus offset are weighted and summed to obtain a comprehensive superposition value, and index matching is performed on a preset threshold interval table based on the comprehensive superposition value, and the matched basic threshold segment is range-scaled to determine the user's skin tolerance threshold interval.

[0098] The product data management platform accesses a product database stored on a local server or in the cloud, retrieves a list of candidate beauty products to be recommended, and uses an optical character recognition engine to scan the ingredient list on the product packaging or instruction manual, converting it into editable text data. Candidate beauty products can include categories such as toners, lotions, serums, and creams.

[0099] Step S104: Analyze and quantify the ingredient text data, separate the chemical component characteristics and concentration characteristics, and calculate the irritation value of the candidate beauty products based on the chemical component characteristics and concentration characteristics to obtain the sensitization risk characteristics.

[0100] In the embodiments of this application, a bidirectional long short-term memory network and a conditional random field model are used to perform sequence labeling and entity extraction on the component text data, identify component entities such as alcohol, flavoring, preservatives and surfactants, and use regular expressions to match the numerical information in the text to extract the concentration percentage and permeation rate coefficient of each component and construct a concentration feature vector.

[0101] The component entity sequence is linked to a pre-defined chemical substance knowledge graph, and the chemical properties, risk labels, and allergen numbers of each component are extracted to generate a chemical component feature vector.

[0102] The proportion of alcohol content and the strength of surfactant in the chemical composition characteristics are extracted, multiplied by preset weights and added together to obtain the superimposed weight. The residual amount of preservative and the amount of heavy metal leaching in the concentration characteristics are accumulated to obtain the cumulative risk value. The superimposed weight and the cumulative risk value are substituted into the linear irritation assessment model for calculation to obtain the irritation value.

[0103] Based on the mapping relationship between irritation values ​​and a pre-defined risk level table, the sensitization risk characteristics of candidate beauty products are determined.

[0104] Step S105: If the sensitization risk characteristics exceed the mapping threshold of the skin tolerance threshold range, and the sensitization markers in the chemical composition characteristics overlap with the user's physiological characteristics, then a micro-conflict is determined, and products with micro-conflicts are eliminated to obtain the remaining products.

[0105] In the embodiments of this application, the upper limit of the skin tolerance threshold range is extracted as the mapping threshold, and the irritation value is compared with the mapping threshold. If the irritation value is greater than the mapping threshold, the allergen number sequence in the sensitization risk characteristics is further extracted, and the microvascular fragility is substituted into the spectral mapping function to calculate the sensitive spectral range corresponding to the user.

[0106] Run the set intersection algorithm to determine whether there is an overlap between the allergen number sequence and the sensitive spectrum range; if both conditions are met, set the micro-conflict flag in the status register to true, determine that there is a micro-conflict, and remove the product from the candidate queue; if neither condition is met, retain the product as the remaining product.

[0107] Step S106: Filter the remaining products to obtain a set of safe beauty products, and calculate the degree of fit between the target products in the set of safe beauty products and the microscopic features.

[0108] In the embodiments of this application, the absolute deviation between the pH values ​​of the remaining products and the user's skin pH value is calculated. If the deviation is less than or equal to a preset pH tolerance threshold, the products are retained; otherwise, they are discarded, resulting in a set of safe beauty products. Based on the product data management method, according to the sensitive skin identifiers and allergen information in the skin characteristic dataset, the beauty product database is traversed to extract the full ingredient data of each product.

[0109] For the target product in the set, the Euclidean distance between its permeation rate coefficient and moisture content index is calculated and normalized to obtain the adsorption fit. The reaction intensity is obtained by calculating the product of the number of fragrance additives and the amount of melanin deposition. The two are weighted and fused to obtain the multimodal feature vector of the target product. The cosine similarity between this vector and the microscopic state feature vector is calculated to obtain the degree of fit. Among them, the physicochemical property deviation may include the pH deviation.

[0110] Step S107: Sort the results according to the matching degree values ​​to obtain the recommendation results.

[0111] In the embodiments of this application, the remaining product list is traversed, the pH value of each remaining product is extracted, and the absolute deviation between the pH value and the epidermal pH value in the microscopic state characteristics is calculated to obtain the pH deviation amount.

[0112] The acid-base deviation is compared with the preset acid-base tolerance threshold. If the acid-base deviation is greater than the acid-base tolerance threshold, the product is deemed unqualified and removed. If the acid-base deviation is not greater than the acid-base tolerance threshold, the product is deemed qualified and included in the safe beauty product collection.

[0113] For each target product in the set of safe beauty products, the Euclidean distance between its penetration rate coefficient and the water content index in the microscopic state characteristics is calculated to obtain the adsorption compatibility. The product of the number of fragrance types added to the target product and the amount of melanin deposition is calculated to obtain the reaction intensity.

[0114] The adsorption fit and reaction intensity are weighted and fused to obtain the multimodal feature vector of the target product. The cosine similarity between the multimodal feature vector and the microstate feature vector is then calculated to obtain the numerical value of the fit.

[0115] Optionally, the fusion weight of adsorption compatibility and reaction intensity can be dynamically adjusted according to the user's skin type.

[0116] The beneficial effects of this application's embodiments compared to traditional technologies are as follows: By extracting multi-dimensional skin physiological parameters through multispectral partitioning sampling to construct microscopic state characteristics, it solves the problem of inaccurate tolerance threshold assessment caused by the single data dimension in traditional methods. A basic fragility index is obtained by analyzing stratum corneum thickness and microvascular fragility, and a personalized skin tolerance threshold range is comprehensively quantified based on the deviation between redness density and epidermal pH value. By parsing and quantifying the ingredient text of candidate products, irritation values ​​are calculated and compared with tolerance thresholds, achieving conflict determination from the microphysiological level to the chemical composition level, avoiding sensitization risks at the source, and significantly improving the safety of recommendations.

[0117] Example 2:

[0118] In an optional embodiment of this application, Figure 2 This is a schematic diagram of Embodiment 2 of the personalized beauty recommendation method based on skin texture detection in this application. Based on microscopic characteristics, the user's skin tolerance threshold range is determined, which may specifically include the following steps:

[0119] Step S201: Call the pre-trained association mapping model, take the stratum corneum thickness and microvascular fragility as the first input group, calculate the nonlinear correlation between the two, and obtain the basic fragility index.

[0120] In the embodiments of this application, the normalized stratum corneum thickness value and the microvascular fragility score are input into a preset nonlinear polynomial regression model, and the basic fragility index is output through the mapping calculation within the model. This index is used to quantify the inherent fragility of the skin's physical barrier and blood vessel walls.

[0121] The pre-trained association mapping model is a nonlinear mapping network based on a multilayer perceptron, and its generation and pre-training process includes:

[0122] A large number of clinical skin test samples were collected. Each sample included labeled stratum corneum thickness, microvascular fragility, red blood vessel density, epidermal pH value, sebum secretion rate, pore size parameters, melanin deposition, water content index (%), and corresponding basic fragility index labeled by dermatologists.

[0123] The model's input layer is an 8-dimensional vector corresponding to 8 skin physiological parameters. There are two hidden layers, each containing 16 neurons, with ReLU activation. The output layer is 1-dimensional, with Sigmoid activation. The Adam optimizer is used, with a preset learning rate of 0.001, a batch size of 32, and mean squared error loss.

[0124] The dataset was divided into training and validation sets in an 8:2 ratio. Iterative training was performed for 500 epochs using backpropagation. The loss was evaluated using the validation set every 50 epochs. Training stopped when the validation loss decreased by less than 0.001 for three consecutive epochs. The model weights were initialized using the He initialization method, and the bias term was initialized to 0.

[0125] Based on the prediction results of the trained model for the entire sample, the probability distribution of the basic vulnerability index is statistically analyzed, and five intervals are divided according to the percentile method: [0, 0.2) is the extremely low vulnerability zone, [0.2, 0.4) is the low vulnerability zone, [0.4, 0.6) is the medium vulnerability zone, [0.6, 0.8) is the high vulnerability zone, and [0.8, 1] is the extremely high vulnerability zone. The corresponding lower limits of the baseline thresholds in the preset threshold interval table are 0.1, 0.3, 0.5, 0.7, and 0.9, and the upper limits are 0.3, 0.5, 0.7, 0.9, and 1.1, respectively. The interval scaling factor is the mean deviation of the tolerance threshold of the samples in each interval.

[0126] In step S202, using the density of red blood vessels and the pH value of the epidermis as the second input group, the absolute value of the difference between the two is calculated to obtain the environmental stimulus offset.

[0127] In the embodiments of this application, the percentage value of the red blood vessel density is calculated as the difference between the epidermal pH value and the standard healthy value (usually 5.0). The absolute value of this difference is taken as the environmental stimulus offset to reflect the potential impact of the external environment on the skin barrier.

[0128] In step S203, using sebum secretion rate and pore size parameters as the third input group, regression analysis is used to determine the erosion coefficient of sebum accumulation on the skin barrier.

[0129] In the embodiments of this application, the sebum secretion rate and pore size parameters are input into a linear regression model, and the erosion coefficient of oil accumulation on the skin barrier is determined based on the slope fitted from historical data. This coefficient characterizes the negative impact of excessive oil secretion on the tolerance threshold.

[0130] Step S204: Using melanin deposition amount and water content as the fourth input group, calculate the attenuation factor of pigment oxidative stress response on tolerance threshold.

[0131] In the embodiments of this application, the ratio of melanin deposition to water content is calculated, and the attenuation factor of pigment oxidative stress response to tolerance threshold is obtained by mapping through a preset exponential function. This factor reflects the reduction in tolerance under the combined effects of pigment deposition and drying.

[0132] Step S205 involves weighted fusion of the basic vulnerability index, environmental stimulus offset, erosion coefficient, and attenuation factor to obtain a comprehensive tolerance index.

[0133] In the embodiments of this application, a comprehensive tolerance index is obtained by weighting and summing four indicators according to a preset weighting strategy, such as a basic vulnerability index weight of 0.4, an environmental stimulus offset weight of 0.2, an erosion coefficient weight of 0.2, and a decay factor weight of 0.2. Specifically, the weighting strategy is based on normalized coefficients obtained by training on historical datasets using machine learning algorithms such as logistic regression or random forests, according to the contribution of each indicator to skin barrier damage. The basic vulnerability index weight is set at 0.4 because it reflects the skin's innate anti-stimulation threshold and is the main controlling factor determining the upper limit of tolerance. The environmental stimulus offset, erosion coefficient, and decay factor weights are all set at 0.2, representing the effects of external inducing factors, chemical damage, and physiological functional degeneration, respectively, and are auxiliary regulatory factors. The weighted summation calculation formula is: Comprehensive tolerance index = 0.4 × basic vulnerability index standardized value + 0.2 × environmental stimulus offset standardized value + 0.2 × erosion coefficient standardized value + 0.2 × decay factor standardized value.

[0134] Step S206: Based on the comprehensive tolerance index, perform index matching and scaling in the preset threshold interval table to obtain the user's skin tolerance threshold interval.

[0135] In the embodiments of this application, a preset threshold range table is queried to find the numerical range in which the comprehensive tolerance index is located, the corresponding lower and upper limits of the benchmark threshold are read, and the benchmark threshold is linearly scaled according to the relative position of the comprehensive tolerance index in the range to output the user's exclusive skin tolerance threshold range.

[0136] In this embodiment, by generating parameters such as stratum corneum thickness and microvascular fragility, red blood vessel density and epidermal pH, sebum secretion rate and pore size, and melanin deposition and water content, the skin tolerance is comprehensively assessed from the perspectives of skin physical barrier fragility, environmental stimuli sensitivity, sebum erosion effect, and pigment oxidation stress, effectively avoiding the one-sidedness of single-parameter assessment. The dynamic tolerance threshold range generated by weighted fusion and interval scaling can accurately reflect the personalized skin tolerance boundary of different users, providing a more scientific reference benchmark for subsequent sensitization risk determination.

[0137] Example 3:

[0138] In an optional embodiment of this application, Figure 3 This is a schematic diagram of Embodiment 3 of the personalized beauty recommendation method based on skin texture detection in this application. The method involves parsing and quantifying ingredient text data to separate chemical component characteristics and concentration characteristics. Specifically, this may include the following steps:

[0139] Step S301: Sequence labeling and entity extraction are performed on the component text data to obtain the component entity sequence. The numerical information in the component text data is parsed to extract the concentration percentage and permeation rate coefficient of each component, and the concentration feature vector is obtained.

[0140] In the embodiments of this application, after obtaining the product ingredient text using an optical character recognition algorithm, sequence labeling is performed using a bidirectional long short-term memory network and a conditional random field model to identify ingredient entities such as "water", "glycerin", and "ethanol".

[0141] By matching numbers and percentage signs in the text using regular expressions, the concentration percentage of each ingredient is extracted. Then, the permeation rate coefficient of each ingredient is estimated using Fick's law diffusion model. All concentration and rate data are then combined sequentially into a concentration feature vector. Specifically, the text of the cosmetic product ingredient list is traversed, detecting values ​​such as "12.5%" and normalizing them to a mass percentage scalar within the range [0,1]. To quantify the actual permeability of these components in skin tissue, this embodiment generates an empirical estimation model based on physiological parameters, rather than solving complex dynamic differential equations. This model utilizes the skin partition coefficient... With skin permeability constant (Determined by the ratio of skin thickness to baseline diffusion coefficient) as the conversion factor. For the first... The component, its effective permeation rate coefficient Calculated using linear combination relationships: This step transforms the originally abstract concentration data into osmotic kinetic indicators with clear physiological significance; finally, according to a predefined order of component importance or chemical formula, the original concentration values ​​are... With the calculated rate coefficient By concatenating them sequentially, a real number sequence vector with a fixed dimension is formed. This serves as the final concentration feature vector.

[0142] Optionally, the conditional random field model is trained using a large amount of labeled cosmetic ingredient text. The ingredient entity sequence may include INCI standard nomenclature and common aliases.

[0143] Step S302: Link the component entity sequence with the preset chemical substance knowledge graph, extract the chemical properties and risk tags of each component, obtain the chemical component feature vector, and integrate the chemical component feature vector and concentration feature vector to obtain the chemical component features and concentration features of the candidate beauty products.

[0144] In the embodiments of this application, the extracted component entity sequence is input into a pre-trained chemical substance knowledge graph embedding model, and the chemical properties, risk tags and allergen numbers corresponding to each component are retrieved and extracted. These discrete features are then subjected to one-hot encoding or word frequency inverse document frequency conversion to generate chemical component feature vectors.

[0145] The chemical component feature vector and the concentration feature vector are concatenated according to their dimensions to form a complete matrix system of chemical component features and concentration features.

[0146] Among them, the chemical substance knowledge graph is a domain knowledge graph for cosmetic ingredients, and its generation process includes:

[0147] It integrates heterogeneous data from multiple sources, including but not limited to PubChem, CosIng, and CTFA; it may also include enterprise-owned data.

[0148] The graph contains three types of nodes, such as chemical substance nodes, risk label nodes, and regulatory nodes; it also contains two types of edge relationships among these three types of nodes, such as having risk and being restricted.

[0149] Component-risk triples are extracted from unstructured documents, synonyms are merged using an entity alignment algorithm, and then knowledge embedding is performed to map nodes into 128-dimensional vectors.

[0150] We crawl the National Medical Products Administration's cosmetic sampling inspection announcements every month, adding nodes for non-compliant ingredients and corresponding risk edges; we update the risk label weights every quarter based on user feedback on allergy cases to keep the knowledge graph timely.

[0151] In this embodiment, a bidirectional long short-term memory network and a conditional random field model are used to achieve sequence labeling and entity extraction of component text. Regular expressions are used to match concentration values, which solves the problems of low efficiency and error-proneness in traditional manual analysis of component text. By linking chemical knowledge graphs to extract chemical properties and risk tags of components, the efficient conversion of unstructured text data into structured feature vectors is achieved, providing a standardized data foundation for subsequent stimulus numerical calculations and risk assessment, and improving the accuracy of component analysis.

[0152] Example 4:

[0153] In an optional embodiment of this application, Figure 4 This is a schematic diagram of Embodiment 4 of the personalized beauty recommendation method based on skin texture detection in this application. Based on the characteristics of chemical components and concentration, the irritation value of candidate beauty products is calculated to obtain sensitization risk characteristics. Specifically, this may include the following steps:

[0154] Step S401: Extract the alcohol content ratio and surfactant strength from the chemical component characteristics.

[0155] In the embodiments of this application, the dimension value representing the alcohol content is indexed from the chemical component feature vector, and the hydrophilic-lipophilic balance value of the corresponding component is queried from an external surfactant database as the surfactant strength.

[0156] Step S402: Calculate the weighted average of alcohol content ratio and surfactant strength, and sum the residual amount of preservative and the amount of heavy metal leaching in the concentration characteristics to obtain the cumulative risk value. The weighted average is calculated as follows: weighted average = alcohol content ratio × corresponding weight of alcohol + surfactant strength × corresponding weight of surface activity. When the alcohol content ratio is greater than the preset ratio, the corresponding weight of alcohol is the first weight coefficient; otherwise, the corresponding weight of alcohol is the second weight coefficient. When the surfactant strength is greater than the preset strength, the corresponding weight of surface activity is the third weight coefficient; otherwise, the corresponding weight of surface activity is the fourth weight coefficient.

[0157] In the embodiments of this application, the preset high alcohol content threshold is 10% (i.e., 0.1). When the alcohol content is ≥10%, the weight corresponding to alcohol is the first weight coefficient of 0.6, otherwise the second weight coefficient of 0.2 is used. The preset high activity threshold is HLB value ≥12. When the surfactant strength is ≥12, the weight corresponding to surface activity is the third weight coefficient of 0.004, otherwise the fourth weight coefficient of 0.001 is used.

[0158] The weighting is calculated through a linear combination. For example, when the alcohol content is 15% (i.e., 0.15) and the HLB value is 13, the weighting is 0.15 × 0.6 + 13 × 0.004 = 0.142. Simultaneously, the residual preservative and leached heavy metal amounts from the concentration characteristics are added together. For example, when the residual preservative is 0.1 and the leached heavy metal is 0.02, the cumulative risk value is 0.1 + 0.02 = 0.12.

[0159] Step S403: Determine the stimulus value based on the superposition weight and the cumulative risk value.

[0160] In the embodiments of this application, the superimposed weight and the cumulative risk value are added together and multiplied by a constant coefficient, and the irritant value of a specific chemical component is calculated by a linear formula.

[0161] Step S404: Based on the mapping relationship between irritation values ​​and a preset risk level table, determine the sensitization risk characteristics of candidate cosmetic products. The risk level table is divided into low risk, medium risk, high risk, and very high risk in ascending order of irritation values. Low risk, medium risk, high risk, and very high risk correspond to differentiated risk warning strategies.

[0162] In the embodiments of this application, for low-risk products falling within the [0, 0.3) range, an allow recommendation strategy is implemented, directly including them in the recommendation list without displaying additional warnings; for medium-risk products in the [0.3, 0.6) range, a yellow label is added to the details page and an allergy test prompt is forcibly displayed; for high-risk products in the [0.6, 0.9) range, a red prohibition icon is displayed on the front end to restrict visibility to users with sensitive skin, and their recommendation weight is reduced in the background; for extremely high-risk products in the [0.9, 1.0] range, the product is permanently removed from the candidate pool and is not recommended or displayed in any way.

[0163] In this embodiment, the synergistic destructive power of solvents and detergents is quantified by calculating the weighted sum of alcohol content ratio and surfactant strength. At the same time, the cumulative risk value of preservative residue and heavy metal leaching is combined to achieve a multi-dimensional comprehensive assessment of the irritant properties of chemical components. The sensitization risk characteristics are determined by the mapping relationship of a preset risk level table, avoiding the limitations of single-component risk assessment.

[0164] Example 5:

[0165] In an optional embodiment of this application, Figure 5 This is a schematic diagram of Embodiment 5 of the personalized beauty recommendation method based on skin texture detection in this application. Calculating the degree of fit between the target product in the safe beauty product set and its microscopic features can specifically include the following steps:

[0166] Step S501: Calculate the Euclidean distance between the permeation rate coefficient and the moisture content index of the target product to obtain the adsorption compatibility, and calculate the product between the number of fragrance additives and the amount of melanin deposition in the target product to obtain the reaction intensity.

[0167] In the embodiments of this application, the permeation rate coefficient and the water content index in the microscopic state characteristics of the target product are read, the Euclidean distance between the two in the characteristic space is calculated, and converted into adsorption fit through an inverse proportional normalization function; at the same time, the number of fragrance types added in the target product formula is counted, and the reaction intensity is obtained by multiplying it by the amount of melanin deposition.

[0168] Step S502: Combine adsorption fit and reaction intensity to obtain the multimodal feature vector of the target product.

[0169] In the embodiments of this application, adsorption compatibility and reaction intensity are used as two dimensions and combined to construct a multimodal feature vector of the target product.

[0170] Step S503: Calculate the cosine similarity between the multimodal feature vector and the microstate feature vector to obtain the numerical value of the degree of fit.

[0171] In the embodiments of this application, the cosine value of the angle between the multimodal feature vector and the micro-state feature vector including the thickness of the stratum corneum, sebum secretion rate, etc. is calculated, and the cosine value is mapped and amplified to the range of 0-100 to obtain the degree of fit value.

[0172] In this embodiment, the adsorption compatibility is calculated by the Euclidean distance between the permeation rate coefficient and the moisture content index, quantifying the degree of matching between the product's permeation performance and the skin condition; the correlation between fragrance ingredients and pigmentation risk is considered by combining the reaction intensity calculation of the types of fragrance added and the amount of melanin deposition; and the fit is calculated by the cosine similarity between multimodal feature vectors and microscopic features, realizing a comprehensive matching evaluation of the product in both permeation compatibility and pigmentation reaction dimensions, thereby improving the deep fit between the recommended product and the user's skin type.

[0173] Example 6:

[0174] In an optional embodiment of this application, Figure 6 This is a schematic diagram of Embodiment Six of the personalized beauty recommendation method based on skin texture detection in this application. Filtering the remaining products may specifically include the following steps:

[0175] Step S601: Extract the pH value of the remaining product and calculate the absolute deviation between the pH value and the pH value of the skin to obtain the pH deviation amount.

[0176] In the embodiments of this application, the calibrated pH value is read from the attribute label of the remaining product, and the user's skin pH value is retrieved from the microscopic state characteristics. The absolute value of the difference between the two is calculated, i.e.: The pH deviation is obtained. Specifically, if the pH value of a product is 6.5 and the pH value of the user's skin is 5.0, the pH deviation is 1.5; if the pH value of the user's skin is 5.8, the deviation is 0.7.

[0177] Step S602: Determine whether the acid-base deviation is greater than the preset acid-base tolerance threshold, wherein the acid-base tolerance threshold is obtained by dynamically correcting the base threshold and the skin pH value.

[0178] In the embodiments of this application, the calculated pH deviation is numerically compared with the dynamically corrected pH tolerance threshold. Specifically, the design logic of the dynamic correction formula is based on the physiological common sense that the pH value of healthy skin is usually maintained between 4.5 and 6.5. The baseline threshold of 0.5 represents the optimal tolerance range for neutral skin. When the user's epidermal pH value deviates from the healthy baseline value of 5.0, the skin's buffering capacity decreases, so the tolerance threshold needs to be narrowed accordingly.

[0179] The dynamic correction formula is: acid-base tolerance threshold = 0.5 - |skin pH value - 5.0| × 0.2. If the user's skin pH value is 5.0, the corrected threshold is 0.5; if the user's pH value is 5.8 (deviation from 0.8), the corrected threshold is 0.5 - 0.8 × 0.2 = 0.34.

[0180] In step S603, if the acid-base deviation exceeds the acid-base tolerance threshold, the remaining products are deemed unqualified and are removed.

[0181] In the embodiments of this application, if the comparison result shows that the pH deviation is greater than the corrected threshold, it is determined that the product's pH environment is incompatible with the user's skin physiological environment. Forced use may disrupt the skin barrier's pH balance, causing stinging or inflammatory reactions. Specifically, if the user's pH value is 5.8 and the tolerance threshold is 0.34, while a product's pH value is 6.5, the deviation is 0.7 > 0.34, then the product is deemed unqualified and rejected.

[0182] In step S604, if the acid-base deviation is not greater than the acid-base tolerance threshold, the remaining products are deemed qualified and included in the safe beauty product set.

[0183] In the embodiments of this application, if the comparison result shows that the acid-base deviation is not greater than the corrected threshold, the product's physicochemical properties are determined to be compatible with the user's skin physiological environment and will not cause acid-base stress. Specifically, if the user's pH value is 5.0 and the tolerance threshold is 0.5, and a product's pH value is 5.4, with a deviation of 0.4 < 0.5, then the product is determined to be qualified and included in the safe beauty product set.

[0184] In this embodiment, the absolute deviation between the pH value and the epidermal pH value is calculated to achieve compatibility filtering of the product's physicochemical properties and the skin's physiological environment. Through the threshold judgment mechanism with a preset acid-base tolerance threshold, products that may disrupt the skin's acid-base balance can be quickly eliminated, further reducing the risk of irritation after product use. This, together with the micro-conflict elimination mechanism, forms a dual safety guarantee, effectively improving the overall safety of the recommended product set.

[0185] Example 7:

[0186] In an optional embodiment of this application, Figure 7 This is a schematic diagram of Embodiment Seven of the Personalized Beauty Product Recommendation Method Based on Skin Type Detection in this application. The remaining products are filtered to obtain a safe beauty product set, which may specifically include the following steps:

[0187] Step S701: Assess the user's skin sensitivity level based on the density of red blood vessels and the fragility of microvessels, and obtain the formula mildness rating and allergen exclusion label for the remaining products.

[0188] In the embodiments of this application, users are divided into three sensitivity levels based on the numerical combination of red blood vessel density and microvascular fragility. Level 1 sensitivity corresponds to a red blood vessel density higher than the first density threshold and a microvascular fragility score greater than the first fragility threshold, indicating that the skin barrier is extremely fragile. Level 2 sensitivity corresponds to a red blood vessel density between the first and second density thresholds, or a microvascular fragility score between the first and second fragility thresholds, indicating that the skin has a significant risk of stress. Insensitivity corresponds to a situation where both indicators are within the normal baseline range.

[0189] The formula mildness rating is a 1-5 level label pre-stored in the product database, derived from data from chicken embryo chorioallantoic membrane testing or skin patch testing; the all-allergen exclusion label indicates that the product formula excludes all known common allergens; the preset sensitivity threshold is uniformly set to level 4, meaning that only products with a mildness rating ≥ 4 are allowed to pass the secondary sensitivity screening.

[0190] Step S702: If the skin sensitivity level is Level 1, only products with the highest formula mildness rating and labeled with all allergen exclusion will be retained to obtain a safe collection of beauty products.

[0191] In the embodiments of this application, the strictest access control is implemented for Level 1 sensitive users, and only products with a formula mildness rating of Level 5 (the highest level) and labeled with a complete allergen exclusion mark are retained.

[0192] Step S703: If the skin sensitivity level is level 2, products with a formula mildness rating lower than the preset sensitivity threshold or without a label indicating exclusion of all allergens are excluded to obtain a set of safe beauty products.

[0193] In the implementation of this application, for users with level 2 sensitivity, two types of products are excluded: products with a formula mildness rating of <4 and products that do not have a label indicating exclusion of all allergens. Only products that meet both requirements are retained.

[0194] Step S704: If the skin sensitivity level is non-sensitive, retain the remaining products as a safe beauty product collection.

[0195] In the embodiments of this application, for non-sensitive users, since their skin tolerance is good, there are no additional restrictions on mildness and allergen labeling. They only need to pass the basic pH filter of claim 6 (i.e., pH deviation ≤ preset pH tolerance threshold) to be classified into the safe beauty product set.

[0196] In this embodiment, sensitivity levels are assessed based on the density of red blood vessels and the fragility of microvessels, achieving a tiered safety filter. This effectively protects highly sensitive users while preventing normal skin users from being overly restricted.

[0197] Example 8:

[0198] In an optional embodiment of this application, Figure 8 This is a schematic diagram of Embodiment 8 of the personalized beauty recommendation method based on skin texture detection in this application. The reaction intensity is obtained by calculating the product between the number of fragrance types added to the target product and the amount of melanin deposition. Specifically, it may include the following steps:

[0199] Step S801: If the amount of melanin deposition is greater than the preset deposition threshold, then assign a first weighting coefficient to the number of fragrance additives.

[0200] In the embodiments of this application, it is determined whether the user's melanin deposition amount reaches or exceeds a preset deposition threshold, such as 100 micrograms per square millimeter. If so, it indicates that the user's skin color is dark or there are pigmentation spots. At this time, a first weighting coefficient with a higher number of fragrance additives is assigned, such as 40, to amplify the influence weight of fragrance on pigment.

[0201] Step S802: If the amount of melanin deposition is not greater than the deposition threshold, then assign a second weighting coefficient to the number of fragrance additions, wherein the second weighting coefficient is less than the first weighting coefficient.

[0202] In the embodiments of this application, if the user's melanin deposition is lower than a preset deposition threshold, it indicates that the user's skin tone is light. In this case, a second weighting coefficient with a lower number of fragrance additives is assigned, such as 20, to reduce the weight of the fragrance's influence on pigmentation.

[0203] Step S803: The reaction intensity is obtained by weighting the number of flavoring types added according to the assigned weighting coefficients.

[0204] In the embodiments of this application, the number of flavoring types added is multiplied by the weighting coefficient determined in the corresponding step to obtain the final reaction intensity value.

[0205] In this embodiment, by dynamically assigning different weighting coefficients to different types of fragrance additives based on the comparison results of melanin deposition amount and preset deposition threshold, a personalized assessment of the reaction intensity of fragrance components and pigments is achieved; the reaction intensity calculation logic is adjusted according to the differentiated needs of users with different skin tones, avoiding the deviation of fixed weight calculation, and further improving the accuracy and personalization level of the fit assessment.

[0206] Example 9:

[0207] In an optional embodiment of this application, Figure 9 This is a schematic diagram of Embodiment Nine of the personalized beauty recommendation method based on skin texture detection in this application. The recommendation results are obtained by sorting the results according to the degree of fit. Specifically, it may include the following steps:

[0208] Step S901: Sort the results according to the degree of fit to obtain the initial sorting result.

[0209] In the embodiments of this application, the compatibility value of each product in the safe beauty product set is read, and the quicksort algorithm is used to sort these values ​​in descending order to generate an initial sorting result list arranged from high to low compatibility.

[0210] Step S902: Extract a preset number of candidate products that rank high in the initial sorting list, and obtain the preset applicable parameters of the candidate products. The preset applicable parameters include oil control efficacy level and pore refinement level.

[0211] In the embodiments of this application, the top N products with the highest degree of fit are selected from the initial sorting list as candidate products, and the oil control efficacy level and pore-refining level parameters of these candidate products are read from the product database.

[0212] Step S903: Based on the sebum secretion rate, pore size parameters, and preset applicable parameters, the corresponding matching level is obtained.

[0213] In the embodiments of this application, the user's sebum secretion rate, pore size parameters, and the oil control efficacy level and pore modification level of the candidate products are input into the trained random forest adaptation model. The model calculates the cosine similarity between feature vectors to match the most suitable adaptation level for each candidate product.

[0214] In this embodiment, highly matched candidate products are selected by sorting by fit, and the matching level is achieved by combining the product's preset applicable parameters with the user's sebum secretion rate and pore size parameters. This achieves a refined transition. The structured recommendation results output specific products and their matching levels, providing users with directly applicable usage guidance and improving the practicality of the recommendation results.

[0215] Figure 10 This illustration shows a structural diagram of a personalized beauty recommendation system based on skin texture detection provided in an embodiment of this application. Specifically, the aforementioned personalized beauty recommendation system 1000 based on skin texture detection may include:

[0216] The acquisition module 1001 is used to acquire the user's facial skin image;

[0217] The sampling module 1002 is used to perform multispectral partition sampling on facial skin images and extract multi-dimensional skin physiological parameters as the user's micro-state features. Among them, the micro-state features include stratum corneum thickness, sebum secretion rate, red blood vessel density, pore size parameters, melanin deposition amount, water content index, microvascular fragility and epidermal pH value.

[0218] The determination module 1003 is used to determine the user's skin tolerance threshold range based on microscopic state characteristics and to obtain the ingredient text data of candidate beauty products. The skin tolerance threshold range includes the basic fragility index and the environmental stimulus offset.

[0219] The calculation module 1004 is used to parse and quantify the ingredient text data, separate the chemical component characteristics and concentration characteristics, and calculate the irritation value of the candidate beauty products based on the chemical component characteristics and concentration characteristics to obtain the sensitization risk characteristics.

[0220] The elimination module 1005 is used to determine that there is a micro-conflict if the sensitization risk characteristics exceed the mapping threshold of the skin tolerance threshold range and the sensitization label in the chemical composition characteristics overlaps with the user's physiological characteristics, and to eliminate the products with micro-conflicts to obtain the remaining products.

[0221] The filtering module 1006 is used to filter the remaining products to obtain a set of safe beauty products, and to calculate the degree of fit between the target products in the set of safe beauty products and the microscopic features.

[0222] The recommendation module 1007 is used to sort the results based on the degree of relevance to obtain the recommendation results.

[0223] The beneficial effects of this application's embodiments compared to traditional technologies are as follows: By extracting multi-dimensional skin physiological parameters through multispectral partitioning sampling to construct microscopic state characteristics, it solves the problem of inaccurate tolerance threshold assessment caused by the single data dimension in traditional methods. A basic fragility index is obtained by analyzing stratum corneum thickness and microvascular fragility, and a personalized skin tolerance threshold range is comprehensively quantified based on the deviation between redness density and epidermal pH value. By parsing and quantifying the ingredient text of candidate products, irritation values ​​are calculated and compared with tolerance thresholds, achieving conflict determination from the microphysiological level to the chemical composition level, avoiding sensitization risks at the source, and significantly improving the safety of recommendations.

[0224] Module 1003 can also be specifically used for: calling a pre-trained association mapping model, using stratum corneum thickness and microvascular fragility as the first input group, calculating the nonlinear correlation between the two to obtain the basic fragility index; using red blood vessel density and epidermal pH value as the second input group, calculating the absolute value of the difference between the two to obtain the environmental stimulus offset; using sebum secretion rate and pore size parameters as the third input group, determining the erosion coefficient of oil accumulation on the skin barrier through regression analysis; using melanin deposition amount and water content as the fourth input group, calculating the attenuation factor of pigment oxidative stress response on the tolerance threshold; weightedly fusing the basic fragility index, environmental stimulus offset, erosion coefficient, and attenuation factor to obtain a comprehensive tolerance index; and indexing and scaling the comprehensive tolerance index in a preset threshold interval table to obtain the user's skin tolerance threshold interval.

[0225] The calculation module 1004 can also be specifically used for: performing sequence labeling and entity extraction on ingredient text data to obtain ingredient entity sequences, parsing the numerical information in the ingredient text data, extracting the concentration percentage and penetration rate coefficient of each ingredient, and obtaining a concentration feature vector; linking the ingredient entity sequences with a preset chemical substance knowledge graph, extracting the chemical properties and risk tags of each ingredient, obtaining a chemical component feature vector, and integrating the chemical component feature vector and the concentration feature vector to obtain the chemical component features and concentration features of candidate cosmetic products.

[0226] The calculation module 1004 can also be specifically used for: extracting the alcohol content ratio and surfactant strength from the chemical component characteristics; calculating the superimposed weight of the alcohol content ratio and surfactant strength, and accumulating the preservative residue and heavy metal leaching in the concentration characteristics to obtain the cumulative risk value. The superimposed weight is calculated as: alcohol content ratio × alcohol corresponding weight + surfactant strength × surface activity corresponding weight. When the alcohol content ratio is greater than a preset ratio, the alcohol corresponding weight takes the first weight coefficient; otherwise, the alcohol corresponding weight takes the second weight coefficient. When the surfactant strength is greater than a preset strength, the surface activity corresponding weight takes the third weight coefficient; otherwise, the surface activity corresponding weight takes the fourth weight coefficient. Based on the superimposed weight and the cumulative risk value, the irritation value is determined. Based on the mapping relationship of the irritation value in a preset risk level table, the sensitization risk characteristics of candidate cosmetic products are determined. The risk level table is divided into low risk, medium risk, high risk, and very high risk in ascending order of irritation value, with low risk, medium risk, high risk, and very high risk corresponding to differentiated risk warning strategies.

[0227] The calculation module 1004 can also be specifically used to: calculate the Euclidean distance between the permeation rate coefficient and the moisture content index of the target product to obtain the adsorption fit; calculate the product between the number of fragrance additives and the amount of melanin deposition in the target product to obtain the reaction intensity; fuse the adsorption fit and the reaction intensity to obtain the multimodal feature vector of the target product; and calculate the cosine similarity between the multimodal feature vector and the microstate feature vector to obtain the degree of fit value.

[0228] The filter module 1006 can also be specifically used for: extracting the pH value of the remaining products and calculating the absolute deviation between the pH value and the skin pH value to obtain the pH deviation amount; determining whether the pH deviation amount is greater than a preset pH tolerance threshold, wherein the pH tolerance threshold is dynamically corrected from the base threshold and the skin pH value; if the pH deviation amount is greater than the pH tolerance threshold, the remaining products are determined to be unqualified and removed; if the pH deviation amount is not greater than the pH tolerance threshold, the remaining products are determined to be qualified and included in the safe beauty product set.

[0229] The filter module 1006 can also be specifically used to: assess a user's skin sensitivity level based on the density of red blood vessels and the fragility of microvessels, and obtain the formula mildness rating and allergen exclusion label of the remaining products; if the skin sensitivity level is Level 1, only products with the highest formula mildness rating and labeled with all allergen exclusion will be retained to obtain a set of safe beauty products; if the skin sensitivity level is Level 2, products with a formula mildness rating below the preset sensitivity threshold or without labeled with all allergen exclusion will be removed to obtain a set of safe beauty products; if the skin sensitivity level is not sensitive, the remaining products will be retained as a set of safe beauty products.

[0230] The filter module 1006 can also be specifically used to: if the amount of melanin deposition is greater than a preset deposition threshold, assign a first weighting coefficient to the number of fragrance additives; if the amount of melanin deposition is not greater than the deposition threshold, assign a second weighting coefficient to the number of fragrance additives, wherein the second weighting coefficient is less than the first weighting coefficient; and calculate the reaction intensity by weighting the number of fragrance additives according to the assigned weighting coefficients.

[0231] The recommendation module 1007 can also be used to: sort products according to their compatibility scores to obtain an initial sorting result; extract a preset number of candidate products from the top of the initial sorting list and obtain the preset applicable parameters of the candidate products, including the oil control efficacy level and the pore refinement level; and match the corresponding suitability level according to the sebum secretion rate, pore size parameters, and preset applicable parameters.

[0232] It should be noted that, in this document, the terms "comprising," "including," and any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Specific examples have been used in this document to illustrate the principles and implementation methods of the technical solutions of this application. The above examples are only for the purpose of helping to understand the methods and core ideas of this application. The above are merely preferred embodiments of this application. It should be pointed out that, due to the limitations of written expression and the objective existence of infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner; these improvements, modifications, changes, or combinations, or the direct application of the concept and technical solutions of this application to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A personalized beauty recommendation method based on skin texture analysis, characterized in that, The method includes: Obtain the user's facial skin image; Multispectral partitioning sampling is performed on the facial skin image to extract multidimensional skin physiological parameters as the user's micro-state features. The micro-state features include stratum corneum thickness, sebum secretion rate, red blood vessel density, pore size parameters, melanin deposition, water content index, microvascular fragility, and epidermal pH value. Based on the microscopic characteristics, the user's skin tolerance threshold range is determined, and the ingredient text data of candidate beauty products is obtained. The ingredient text data is parsed and quantified to separate chemical component characteristics and concentration characteristics. Based on the chemical component characteristics and concentration characteristics, the irritation value of the candidate cosmetic product is calculated to obtain the sensitization risk characteristics. If the sensitization risk characteristics exceed the mapping threshold of the skin tolerance threshold range, and the sensitization markers in the chemical composition characteristics overlap with the user's physiological characteristics, then a micro-conflict is determined, and products with micro-conflicts are eliminated to obtain the remaining products. The remaining products are filtered to obtain a set of safe beauty products, and the degree of fit between the target products in the set of safe beauty products and the microscopic features is calculated. The results are sorted according to the degree of fit to obtain the recommendation results.

2. The personalized beauty recommendation method based on skin texture detection as described in claim 1, characterized in that, Determining the user's skin tolerance threshold range based on the microscopic characteristics includes: The pre-trained association mapping model is invoked, and the stratum corneum thickness and the microvascular fragility are used as the first input group to calculate the nonlinear correlation between the two and obtain the basic fragility index. Using the density of red blood vessels and the pH value of the epidermis as the second input group, the absolute value of the difference between the two is calculated to obtain the environmental stimulus offset. Using the sebum secretion rate and the pore size parameter as the third input group, regression analysis was used to determine the erosion coefficient of oil accumulation on the skin barrier. Using the melanin deposition amount and the water content index as the fourth input group, the attenuation factor of pigment oxidative stress response to tolerance threshold is calculated. The basic vulnerability index, the environmental stimulus offset, the erosion coefficient, and the attenuation factor are weighted and fused to obtain a comprehensive tolerance index. The user's skin tolerance threshold range is obtained by indexing and scaling the comprehensive tolerance index in a preset threshold range table.

3. The personalized beauty recommendation method based on skin texture detection as described in claim 1, characterized in that, The process of parsing and quantifying the component text data to separate chemical component features and concentration features includes: Sequence labeling and entity extraction are performed on the component text data to obtain the component entity sequence. The numerical information in the component text data is parsed to extract the concentration percentage and permeation rate coefficient of each component, and a concentration feature vector is obtained. The component entity sequence is linked with a preset chemical substance knowledge graph to extract the chemical properties and risk tags of each component, thereby obtaining a chemical component feature vector. The chemical component feature vector is then integrated with the concentration feature vector to obtain the chemical component features and concentration features of the candidate cosmetic product.

4. The personalized beauty recommendation method based on skin texture detection as described in claim 1, characterized in that, The step of calculating the irritation value of the candidate cosmetic product based on the chemical composition characteristics and the concentration characteristics to obtain the sensitization risk characteristics includes: Extract the alcohol content ratio and surfactant strength from the characteristics of the chemical components; The cumulative risk value is obtained by calculating the weighted sum of the alcohol content ratio and the surfactant strength, and summing the preservative residue and heavy metal leaching in the concentration characteristics. The weighted sum is calculated as follows: alcohol content ratio × alcohol corresponding weight + surfactant strength × surface activity corresponding weight. When the alcohol content ratio is greater than a preset ratio, the alcohol corresponding weight is taken as the first weight coefficient; otherwise, the alcohol corresponding weight is taken as the second weight coefficient. When the surfactant strength is greater than a preset strength, the surface activity corresponding weight is taken as the third weight coefficient; otherwise, the surface activity corresponding weight is taken as the fourth weight coefficient. The stimulus value is determined based on the superposition weight and the cumulative risk value; Based on the mapping relationship between the irritation value and the preset risk level table, the sensitization risk characteristics of the candidate cosmetic products are determined. The risk level table is divided into low risk, medium risk, high risk, and very high risk in ascending order of irritation value. The low risk, medium risk, high risk, and very high risk correspond to differentiated risk warning strategies.

5. The personalized beauty recommendation method based on skin texture detection as described in claim 4, characterized in that, The calculation of the degree of fit between the target product in the set of safe beauty products and the microscopic features includes: The Euclidean distance between the permeation rate coefficient of the target product and the moisture content index is calculated to obtain the adsorption compatibility. The product between the number of fragrance additives in the target product and the amount of melanin deposition is calculated to obtain the reaction intensity. By combining the adsorption adaptability and the reaction intensity, the multimodal feature vector of the target product is obtained; Calculate the cosine similarity between the multimodal feature vector and the microstate feature vector to obtain a numerical value of the degree of fit.

6. The personalized beauty recommendation method based on skin texture detection as described in claim 1, characterized in that, The filtering of the remaining products includes: Extract the pH value of the remaining product and calculate the absolute deviation between the pH value and the pH value of the skin to obtain the pH deviation amount; Determine whether the acid-base deviation is greater than a preset acid-base tolerance threshold, wherein the acid-base tolerance threshold is obtained by dynamically correcting the base threshold and the epidermal pH value; If the acid-base deviation exceeds the acid-base tolerance threshold, the remaining products are deemed unqualified and discarded. If the acid-base deviation is not greater than the acid-base tolerance threshold, the remaining products are deemed qualified and included in the safe beauty product set.

7. The personalized beauty recommendation method based on skin texture detection as described in claim 1, characterized in that, The remaining products are filtered to obtain a set of safe beauty products, including: Based on the density of red blood vessels and the fragility of microvessels, assess the user's skin sensitivity level and obtain the formula mildness rating and allergen exclusion label of the remaining products; If the skin sensitivity level is Level 1, then only products with the highest formula mildness rating and labeled with the all-allergen exclusion mark will be retained to obtain the safe beauty product set. If the skin sensitivity level is level 2, products with a formula mildness rating lower than the preset sensitivity threshold or without the all-allergen exclusion label are excluded to obtain the set of safe beauty products. If the skin sensitivity level is "not sensitive", then the remaining products are retained as part of the safe beauty product set.

8. The personalized beauty recommendation method based on skin texture detection as described in claim 7, characterized in that, The calculation of the product between the number of fragrance types added to the target product and the amount of melanin deposition to obtain the reaction intensity includes: If the amount of melanin deposited is greater than a preset deposition threshold, a first weighting coefficient is assigned to the number of fragrance additives. If the amount of melanin deposition is not greater than the deposition threshold, then a second weighting coefficient is assigned to the number of fragrance additives, wherein the second weighting coefficient is less than the first weighting coefficient; The reaction intensity is obtained by weighting the number of flavoring types added according to the assigned weighting coefficients.

9. The personalized beauty recommendation method based on skin texture detection as described in claim 1, characterized in that, The process of sorting the results based on the matching scores to obtain recommendations includes: The initial sorting result is obtained by sorting the results according to the compatibility scores. Extract a preset number of candidate products that rank highly in the initial sorting list, and obtain preset applicable parameters for the candidate products, including oil control efficacy level and pore minimization level. Based on the sebum secretion rate, the pore size parameter, and the preset applicable parameters, a corresponding matching level is obtained.

10. A personalized beauty recommendation system based on skin texture detection, characterized in that, include: The acquisition module is used to acquire the user's facial skin image; The sampling module is used to perform multispectral partitioning sampling on the facial skin image and extract multi-dimensional skin physiological parameters as the user's micro-state features. The micro-state features include stratum corneum thickness, sebum secretion rate, red blood vessel density, pore size parameters, melanin deposition amount, water content index, microvascular fragility and epidermal pH value. The determination module is used to determine the user's skin tolerance threshold range based on the microscopic state characteristics, and to obtain the ingredient text data of candidate beauty products, wherein the skin tolerance threshold range includes a basic fragility index and an environmental stimulus offset. The calculation module is used to parse and quantify the ingredient text data, separate the chemical component characteristics and concentration characteristics, and calculate the irritation value of the candidate beauty product based on the chemical component characteristics and the concentration characteristics to obtain the sensitization risk characteristics. The elimination module is used to determine that there is a micro-conflict if the sensitization risk characteristics exceed the mapping threshold of the skin tolerance threshold range, and the sensitization marker in the chemical component characteristics overlaps with the user's physiological characteristics, and to eliminate the products with micro-conflicts to obtain the remaining products. A filtering module is used to filter the remaining products to obtain a set of safe beauty products, and to calculate the degree of fit between the target products in the set of safe beauty products and the microscopic features. The recommendation module is used to sort the results according to the matching degree values ​​to obtain recommendation results.

Images