Intelligent device for measuring skin and customizing cosmetics

Abstract

The invention discloses equipment for intelligently measuring skin and customizing cosmetics, and relates to the technical field of intelligent beauty. The device comprises a test unit used for obtaining basic data of a current user; the formula design unit is used for determining respective formulas of various cosmetics suitable for the current user by utilizing a mapping relation model between the basic data and the formulas on the basis of the acquired basic data; the material matching unit is used for matching the entity materials corresponding to the components of the formula into the corresponding component bins; the material taking unit is used for taking out the entity materials in the corresponding proportion and conveying the entity materials to the preparation unit; and the preparation unit is used for preparing cosmetics. Cosmetics prepared by using the equipment provided by the invention can meet the requirement of a user for adjusting a skin care scheme in real time according to skin change, and continuous and accurate skin care is realized. The core value of effective skin care can be fully embodied. In addition, due to skin measurement customization, product components can be well matched with individual skin.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent beauty, and in particular to a device for intelligent skin measurement and customized cosmetics. BACKGROUND

[0002] The development history of the cosmetic industry clearly shows the evolution from "one product for the whole family", to the development of exclusive products for subgroups of people with different skin conditions, to the realization of precise skin care based on functional ingredients.

[0003] The design and production of cosmetics generally have the following defects: 1. Traditional products are difficult to adapt to dynamic skin conditions: Current traditional cosmetics on the market are mostly designed as a complete set, and the use cycle is usually 1-2 months. However, the skin condition fluctuates dynamically with factors such as seasonal changes, environmental changes (such as temperature and humidity, pollution level), etc. The traditional product with fixed formula cannot meet the needs of consumers to adjust the skin care scheme in real time according to the changes in skin condition, and it is difficult to achieve continuous and precise skin care. 2. Packaging costs squeeze the core value: The phenomenon of "emphasizing packaging and ignoring content" is common in the industry. Excessive packaging not only increases the overall cost of the product (the packaging cost accounts for a high proportion of the selling price), but also may cause resource waste, while the quality of raw materials and the effectiveness of the formula, which determine the effectiveness of skin care, are not fully invested, which deviates from the core demand of consumers for "effective skin care". 3. Supply and demand link fault leads to matching imbalance: The circulation link of the existing cosmetic market is long, and there is a clear information barrier between consumers and R&D personnel - consumers cannot directly communicate their skin condition needs to R&D personnel, resulting in insufficient matching between product ingredients and individual skin; at the same time, the feedback from consumers after using the product is difficult to efficiently flow back to the R&D link, and cannot timely drive the iterative optimization of the skin care scheme, breaking the closed loop of "R&D - consumption - feedback". SUMMARY

[0004] In order to solve the problems in the prior art, the present application provides the following technical solutions.

[0005] The first aspect of the present application provides a device for intelligent skin measurement and customized cosmetics, comprising: a test unit for obtaining basic data of a current user, the basic data including skin condition indicators, skin components, and environmental parameters; a formula design unit for determining initial formulas of various types of cosmetics suitable for the current user based on the obtained basic data and using a mapping relationship model between basic data and formulas; and for checking the compliance and applicability of the initial formulas using a knowledge graph associated with cosmetic raw material characteristic data to obtain final formulas; a material matching unit for matching entity materials corresponding to the components of the final formula to corresponding component bins; ingredient bins for storing corresponding physical materials of ingredients; a material taking unit for taking out corresponding physical materials of ingredients according to the final formula from the corresponding ingredient bins and delivering to the preparation unit; a preparation unit for preparing the customized cosmetics by using the physical materials.

[0006] Preferably, the device further comprises a quantitative discharging unit and a cleaning unit; the quantitative discharging unit is used to discharge the prepared product by high-pressure airflow driving; the cleaning unit is used to clean and disinfect the material taking unit, the preparation unit and the quantitative discharging unit.

[0007] Preferably, the testing unit comprises a non-contact skin imaging device, a contact skin component testing device and an environment monitoring device; the non-contact skin imaging device is used to obtain skin texture indexes; the contact skin component testing device is used to obtain skin components; and the environment monitoring device is used to obtain environmental parameters.

[0008] Preferably, the non-contact skin imaging device is used to obtain skin texture indexes, and is implemented in the following manner: Infrared response images and multispectral images of the face are collected; A face skin texture map is generated based on the infrared response images and the multispectral images; A deep learning model is used to obtain skin texture indexes based on the face skin texture map.

[0009] Preferably, the contact skin component testing device is used to obtain skin components, and is implemented in the following manner: The conductivity and the pH value of the skin surface are detected; A pre-constructed mapping relationship between the conductivity and the water content is used to obtain the water content of the skin based on the detected conductivity; A pre-constructed correlation model between the pH value and the grease oxidation degree is used to obtain the grease oxidation degree of the skin based on the detected pH value, and the grease secretion grade is determined based on the grease oxidation degree.

[0010] Preferably, the environment monitoring device is used to obtain environmental parameters, and is implemented in the following manner: Built-in temperature sensors and humidity sensors are used to monitor the temperature and the humidity in real time; An API interface is used to connect to a national meteorological data platform to obtain real-time ultraviolet intensity, PM2.5 and / or wind force data. Preferably, the mapping relationship model between the basic data and the formula is obtained by training in the following manner: Obtaining training data, the training data comprising: basic data, corresponding formula and effect feedback data; Adjusting the skin texture index according to the environmental parameters to obtain an adjusted skin texture index; Determining a skin texture type based on the adjusted skin texture index and skin components; Training a deep reinforcement learning model using the determined skin texture type and the corresponding formula; and updating a reward function of the model using the effect feedback data during the training process.

[0011] Preferably, the adjusting the skin texture index according to the environmental parameters comprises: Determining an importance weight of each environmental parameter on the skin texture index using a multi-random forest model; Determining an influence coefficient of a comprehensive environment on the skin texture index based on the importance weight of each environmental parameter using a multivariate linear regression model; Adjusting the skin texture index based on the influence coefficient to obtain a final skin texture index.

[0012] Preferably, the performing compliance and applicability checking on the initial formula using the knowledge graph associated with cosmetic raw material characteristic data comprises: The entities associated with the knowledge graph include: raw materials, effects, skin texture, environmental factors and cosmetic dosage forms; Based on the knowledge graph, the raw materials and effects are matched through rule-based reasoning and graph algorithms to avoid banned ingredients and adapt to the environment and dosage forms.

[0013] Preferably, the matching the entity materials corresponding to the components of the final formula to the corresponding component warehouses comprises: Converting the component effect requirements of the final formula into characteristic indexes of entity materials; Filtering entity materials with a matching degree ≥ 90% from a raw material database through a cosine similarity algorithm; Evaluating the compatibility between candidate entity materials using a GNN model to exclude combinations with conflicts; When a user manually replaces an entity material, re-perform compatibility verification.

[0014] The beneficial effects of the present application are: the device for intelligent skin measurement and customized cosmetics provided by the present application obtains the skin quality index, skin components and environmental parameters of the current user, and determines the respective formulations of various cosmetics suitable for the current user by using the mapping relationship model between the formulations and the components; the components of the formulations are matched into the corresponding component bins by corresponding the components and the physical materials; the corresponding physical materials are taken out from the corresponding component bins, and the customized cosmetics for the user are prepared in the preparation unit. Compared with the traditional products with fixed formulations, the cosmetics prepared by the device provided by the present application can meet the needs of the user to adjust the skin care scheme in real time according to the changes of skin quality, and realize continuous and accurate skin care. Moreover, there is no need for heavy packaging cost, so that the core value of "effective skin care" can be fully embodied. In addition, the skin measurement and customization of the present application can really start from the actual needs of consumers, the circulation link is short, so that the product components and individual skin can be better matched; at the same time, the effect feedback data of the consumer after using the product can drive the iterative optimization of the skin care scheme, forming a closed loop of "research and development-consumption-feedback". BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 FIG. 1 is a functional structure schematic diagram of the device for intelligent skin measurement and customized cosmetics of the present application; Figure 2 FIG. 2 is a structure schematic diagram of the device for intelligent skin measurement and customized cosmetics of the present application; Figure 2 In the above description, the meanings of various symbols are as follows: 1, non-contact skin imaging device; 2, contact skin component testing device; 3, component bin. DETAILED DESCRIPTION

[0016] In order to better understand the above technical solutions, the above technical solutions will be described in detail in combination with the description of the drawings and the specific embodiments.

[0017] The method provided by the present application can be implemented in a terminal environment, which can include one or more of the following components: a processor, a memory and a display screen. The memory stores at least one instruction, which is loaded and executed by the processor to implement the method described in the following embodiments.

[0018] The processor can include one or more processing cores. The processor connects various parts in the entire terminal through various interfaces and lines, executes various functions of the terminal and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory.

[0019] The memory can include a random access memory (RAM) and can also include a read-only memory (ROM). The memory can be used to store instructions, programs, codes, code sets, or instructions.

[0020] The display screen is used to display the user interface of each application program.

[0021] In addition, those skilled in the art can understand that the structure of the terminal described above does not constitute a limitation on the terminal, and the terminal can include more or fewer components, or combine certain components, or different component arrangements. For example, the terminal also includes radio frequency circuitry, an input unit, a sensor, audio circuitry, a power supply, and the like, which are not described here.

[0022] Embodiment one As shown in the figure, the embodiment of the application provides a device for customizing cosmetics according to skin, which comprises: Figure 1 A test unit 101 is configured to acquire basic data of a current user, the basic data comprising skin texture indicators, skin components, and environmental parameters. The skin texture indicators can include, for example, glossiness, smoothness, pigmentation, sensitivity, and cleanliness. The skin components can include, for example, moisture, oil, and ions. The environmental parameters can include, for example, environmental temperature, humidity, and ultraviolet intensity. Since the above basic data of the user can change at any time and in any place, in the embodiment of the application, the current state of the user and some parameter indicators of the environment are determined, and these parameter indicators are used as a basis for subsequent formula design and cosmetic preparation. A formula design unit 102 is configured to determine initial formulas of various types of cosmetics suitable for the current user based on the acquired basic data, using a mapping relationship model between the basic data and the formulas; and to perform compliance and applicability checks on the initial formulas using a knowledge graph associated with cosmetic raw material characteristic data, to obtain final formulas. The cosmetics can include, for example, cleansing products, makeup water, emulsions, essences, and creams. One or more of them can be included according to requirements. In the embodiment of the application, a model capable of simulating the mapping relationship between the basic data and the formulas is trained to determine the formula suitable for the user based on the acquired basic data of the current user. The final formula is obtained by checking the initial formula using the knowledge graph.

[0023]

[0024] ​The material matching unit 103 is configured to match the entity material corresponding to the ingredient of the final formula to the corresponding ingredient bin. The characteristic data of the entity material can be stored in a material database. In actual application, the ingredient in the formula can be matched to the material corresponding to the ingredient in the formula by retrieving the material database according to the functional requirement of the ingredient in the formula. For each ingredient, an ingredient bin is configured in the device to store the corresponding ingredient material. After the ingredient in the formula is matched to the corresponding material, the required material can be configured to the corresponding ingredient bin, which facilitates the subsequent material taking unit to take the material.

[0025] The ingredient bin 104 is configured to store the entity material corresponding to each ingredient.

[0026] The material taking unit 105 is configured to take out the corresponding proportion of the entity material from the corresponding ingredient bin based on the final formula and deliver the material to the preparation unit. When taking the material, only the material in the corresponding ingredient bin needs to be taken out. The material taking can be completed in an automatic manner.

[0027] The preparation unit 106 is configured to prepare the customized cosmetic for the user by using the entity material. After all the materials are taken out, the target cosmetic is prepared according to a certain process.

[0028] In the embodiment of the present application, the device can further include a quantitative discharging unit and a cleaning unit. The quantitative discharging unit is configured to discharge the prepared product by driving a high-pressure airflow. The cleaning unit is configured to clean and sterilize the material taking unit, the preparation unit and the quantitative discharging unit. The device can complete the complete process of the skin customized cosmetic. Other devices are not required. The device is convenient and fast.

[0029] The cleaning unit can use airflow cleaning. Specifically, a HEPA H13 filter can be used to provide clean air (≥99.97% @ 0.3 μm particles). A UV-C lamp (power 10W, wavelength 254nm) and an ozone generator (concentration 0.1-0.3ppm) are used for sterilization and disinfection. The specific implementation process can be as follows: (1) Pretreatment cleaning: The ingredient bin interface and the sampling needle are blown by clean airflow (10 seconds) before sampling.

[0030] (2) Deep disinfection: After each preparation is completed, the UV + ozone combined disinfection is started (30 seconds), and the residual ozone is decomposed by the activated carbon filter (to ensure that the emission concentration is ≤0.05ppm).

[0031] (3) Periodic maintenance: After 100 times of use, the filter and the UV lamp are automatically prompted to be replaced.

[0032] The cleaning unit provided by the application can achieve the following sterilization effect: the microbial killing rate is greater than or equal to 99.9%, which meets the hygiene standard GB 15979-2002.

[0033] The quantitative discharging unit can be driven by sterile compressed air (0.2-0.4 MPa), and the inner wall of the discharging pipeline is treated by super-hydrophobic coating to reduce material residue. Specifically, the combination of "sterile air source + precision valve body + super-hydrophobic pipeline" can be adopted to ensure the quantitative and residue-free output of the material. The sterile air source module includes: a medical-grade air compressor (output pressure 0.5-0.8 MPa), a multi-stage filter (dust removal / water removal / bacteria removal, filtration precision 0.01 μm) and a pressure reducing valve (adjusted to 0.2-0.4 MPa) to provide clean and stable pressure compressed air. The material cavity connecting assembly can include: a pneumatic stop valve (Φ15 mm, PTFE seal) between the discharging port of the mixing cavity and the storage cavity, the pneumatic stop valve is provided with a position sensor (to confirm the opening and closing state of the valve) and controls the material to enter the discharging buffer cavity from the mixing cavity. The quantitative pushing unit can include a cylindrical storage cavity (volume 50-500 mL, 316L stainless steel), a piston type push rod (food-grade silicone seal ring) and a displacement sensor (accuracy ±0.1 mm) to accurately control the single discharging volume (0.1-100 mL). The super-hydrophobic discharging pipeline can be made of polytetrafluoroethylene (PTFE) pipe (inner diameter Φ6-10 mm), the inner wall of which is sprayed with nano-silicon dioxide super-hydrophobic coating (contact angle ≥150°), and the end is provided with a quick-connect type discharging nozzle to reduce material residue (residual amount ≤0.1 mL / 100 mL).

[0034] The principle of sterile compressed air driving is achieved by indirect means of "pressure transmission-piston pushing", avoiding direct contact between gas and material (to prevent contamination and bubble generation). The specific process is as follows: 1. Air source pretreatment - the compressed air generated by the air compressor is first filtered by three levels: first level: metal mesh filter (remove ≥5μm particulate matter); second level: activated carbon filter (adsorb oil mist, odor); third level: HEPA sterilization filter (filter ≥0.01μm microorganisms, sterilization rate ≥99.99%); the filtered clean air is stabilized to 0.2-0.4MPa by a pressure reducing valve, and stored in a gas storage tank (volume 2L), ensuring that the pressure fluctuation is ≤±0.02MPa. 2. Driving execution process: first step: material filling, the pneumatic stop valve is opened, the material in the mixing chamber flows into the storage chamber under the action of gravity, the displacement sensor detects the position of the piston, and when the preset volume (such as 10mL) is reached, the stop valve is closed, completing the quantitative filling. Second step: air pushing The pneumatic interface at the top of the storage chamber is connected to sterile compressed air (0.3MPa), the gas pressure acts on the back of the piston, pushing the piston to move downward along the inner wall of the storage chamber, and the material is pushed into the outlet pipeline from the bottom. Third step: precise stopping The displacement sensor monitors the piston stroke in real time, and when the pushing volume reaches the set value (such as 10mL), the control system immediately closes the air source electromagnetic valve, and at the same time opens the exhaust valve to release the pressure in the chamber, the piston stops moving, and the quantitative discharge is realized. 3. Anti-residue design, the inner wall of the storage chamber and the piston sealing ring adopt "interference fit" (clearance ≤0.05mm), ensuring that the material does not leak back; the super-hydrophobic coating of the outlet pipeline makes the material form a "rolling effect" (similar to the lotus effect), combined with a 0.5 second reverse airflow blowing (0.2MPa) at the end, the residual amount in the pipeline is controlled within 0.1mL. The connection relationship of the key components is: [medical air compressor] → [three-stage filter] → [pressure reducing valve (0.2-0.4MPa)] → [gas storage tank]; [mixing chamber] → [pneumatic stop valve] → [storage chamber (with piston)] → [super-hydrophobic outlet pipe] → [outlet nozzle]. The control logic is: PLC automatically calculates the required stroke of the piston (through the inner diameter of the storage chamber: stroke = volume / cross-sectional area) according to the set discharge amount of the formula (such as 10mL), and adjusts the air pressure through the pressure sensor closed loop, ensuring that different viscosity materials (such as toner 5cP vs. face cream 5000cP) can be stably pushed. The pressure regulation strategy for different materials is as follows: low viscosity material (such as cosmetic water, ≤100 cP): low pressure driving of 0.2-0.25 MPa is adopted, and slow pushing speed (10 mL / s) is used to avoid material splashing; medium viscosity material (such as emulsion, 100-1000 cP): medium pressure driving of 0.3 MPa is adopted, and the pushing speed is 20 mL / s to balance the efficiency and stability; high viscosity material (such as face cream, >1000 cP): high pressure driving of 0.35-0.4 MPa is adopted, and pipeline heating (30-35 DEG C, controlled by jacket) is started to reduce the viscosity, and the pushing speed is 5 mL / s to ensure complete discharge. Through non-contact driving of sterile air and super-hydrophobic pipeline design, the system can not only meet the hygiene requirements of cosmetic production (in line with GMP standards), but also realize high-precision quantitative discharge (error ≤±1%), while minimizing waste and cross contamination caused by material residue.

[0035] In addition, in the embodiment of the present application, an infrared flowmeter is also provided in the quantitative discharge unit to monitor the discharge amount in real time, with an accuracy of ±0.1 mL.

[0036] Discharge control: according to user requirements (such as 10 mL per time), set the air flow pressure and time, and avoid material splashing through pulse discharge.

[0037] Residual treatment: after the discharge is completed, clean air is introduced into the pipeline in the reverse direction (0.5 MPa, 3 seconds) to ensure that the residual amount is ≤0.1 mL.

[0038] Maximum discharge speed: 10 mL / s, supporting multiple discharges (such as 5 mL in the morning and 5 mL in the evening).

[0039] As shown in FIG. 1, Figure 2 The test unit includes a non-contact skin imaging device 1, a contact skin component testing device 2, and an environment monitoring device (not shown in the figure); wherein the non-contact skin imaging device is used to obtain skin texture indicators; the contact skin component testing device is used to obtain skin components; and the environment monitoring device is used to obtain environmental parameters.

[0040] The non-contact skin imaging device is used to obtain skin texture indicators, which can be implemented in the following manner: Collecting infrared response images and multispectral images of the face; Generating a face skin texture map based on the infrared response images and multispectral images; Using a deep learning model, obtaining skin texture indicators based on the face skin texture map.

[0041] Specifically, 3D multispectral imaging technology can be used, equipped with a high-resolution CMOS sensor (more than 20 million pixels), and combined with multiple LED light sources in the 400-950 nm band (including ultraviolet, visible light, and near-infrared), to realize multi-dimensional imaging of the skin surface and shallow layers. The polarization light filtering module is integrated to eliminate the interference of skin surface reflection and accurately capture details such as pigment deposition and capillary distribution.

[0042] The following process can be used for implementation: User face positioning: trigger imaging through infrared sensing, automatically recognize the face area (forehead, cheek, nose wing, etc.) through built-in AI algorithm, and ensure consistency of the detection area each time. Image acquisition and processing: synchronize multispectral image acquisition, generate complete facial skin quality map through image stitching technology, and then convert image features into quantitative indicators (such as glossiness: 0-100 points, smoothness: texture roughness value μm) through deep learning model (based on million-level skin quality sample training). Data output: generate visual skin quality report, mark pigment deposition area, sensitive point distribution, etc., and upload to background database. Detection accuracy: pigment deposition recognition error ≤5%, texture roughness measurement error ≤2 μm. Detection time: single person ≤30 seconds, supports 3-second fast preview. Deep learning model can realize intelligent analysis and quantitative evaluation of skin images, and accurately extract key indicators of skin quality. Convolutional neural network (CNN) can be used as the core model, combined with attention mechanism, focusing on key areas such as pigment deposition and sensitive points. Migration learning is introduced, and the pre-trained ResNet-50 model is fine-tuned to improve detection accuracy under small samples. The training data covers a million-level skin image dataset of different races, ages, and skin qualities, including annotated glossiness, smoothness, pigment deposition, and other indicator data. Online learning strategy can be used for optimization mechanism, regularly incorporating new user data for model updating, and continuously improving the adaptability to diverse skin quality.

[0043] The specific reasoning process can include: optimizing image quality through denoising, normalization, etc. Use CNN to extract skin texture, color distribution, and other deep features. Map the features to specific quantitative indicators (such as glossiness score, texture roughness value), and simultaneously optimize the prediction accuracy of multiple indicators through multi-task learning.

[0044] In the embodiment of the present application, the contact skin component testing device for obtaining skin components can be implemented in the following manner: detecting the conductivity and PH value of the skin surface; using a pre-constructed mapping relationship between conductivity and water content, obtaining the water content of the skin based on the detected conductivity; using a pre-constructed correlation model between PH value and grease oxidation degree, obtaining the grease oxidation degree of the skin based on the detected PH value, and determining the grease secretion level based on the grease oxidation degree.

[0045] Specifically, a flexible electrode array (including an Ag / AgCl reference electrode) can be used to detect the conductivity (reflecting the water content) and PH value (range 4.0-9.0, accuracy ±0.05) of the skin surface.

[0046] The specific process can include: sensor head disinfection: disinfecting for 10 seconds before each use through an ultraviolet lamp (254 nm) to ensure hygiene and safety. Multi-point detection: automatically selecting three representative areas (such as cheek, t-zone) on the face for contact detection, taking the average value to reduce errors. Data analysis: based on the conductivity-water content calibration curve (established through 3000+ skin sample), calculating the stratum corneum hydration degree (%); through the correlation model between PH value and grease oxidation degree, outputting the grease secretion level (1-5 levels).

[0047] The conductivity-water content calibration curve is used to convert the detected value of the skin conductivity into the quantitative stratum corneum water content (%), and the core is to establish a mathematical mapping relationship between "conductivity-actual water content" through a large number of samples, and the steps are as follows: (1) Sample collection and grouping Sample size: select 3000+ different skin quality volunteers (covering dry, oily, mixed, sensitive skin, age 18-60 years old), to ensure the diversity of samples in gender, age, and region.

[0048] Grouping standard: pre-classified according to initial water content (determined by laboratory standard method), divided into 5 groups: extremely dry (<20%), dry (20%-35%), neutral (35%-50%), oily (50%-65%), and extremely oily (>65%), with 600+ samples in each group to avoid data bias.

[0049] (2) Double method synchronous detection Conductivity detection: using the flexible electrode array of the component analysis system, measuring the conductivity of the face of each volunteer at three fixed areas (cheek, t-zone, mandible) under the same environmental conditions (temperature 25±1℃, humidity 50±5% RH), recording the stable value (unit: μS / cm), and taking the average value of each area for three times.

[0050] Actual moisture content detection: The laboratory gold standard method - Corneometer® CM 825 (Institute for Skin Science certified capacitance moisture meter) is used to measure the stratum corneum moisture content (%) in the same area simultaneously as the "true value".

[0051] (3) Data processing and curve fitting Outlier rejection: 3% of the abnormal data (such as samples with contradictory conductivity and moisture content trends) are removed by boxplot method to ensure data reliability.

[0052] Curve type selection: Draw scatter plots to observe the distribution rules (conductivity increases nonlinearly with increasing moisture content, with high sensitivity at low moisture content and slow growth at high moisture content), choose quadratic polynomial fitting (y = ax² + bx + c) or exponential fitting (y = a e^(bx) + c), which is better than linear fitting (error can be reduced by 40%).

[0053] Parameter calibration: Use least squares method to regress 3000+ sample data to determine the coefficients (a, b, c) of the fitting equation, for example, the fitting result of a certain batch of samples is: Moisture content (%) = 0.002 x conductivity² + 0.35 x conductivity + 5.2 (R²≥0.96, goodness of fit). (4) Verification and optimization Internal verification: Take 20% of the total samples (600+) as the verification set, substitute the curve to calculate the moisture content, and compare it with the laboratory measured value to ensure that the error is ≤3%.

[0054] Dynamic adjustment: Every 500 new samples, re-fit the curve parameters to avoid deviations caused by population differences (such as samples from dry northern regions may need to be corrected separately).

[0055] PH value and grease oxidation degree correlation model is used to predict the degree of grease oxidation (reflecting grease rancidity and skin barrier function) through skin surface PH value. The core is to explore the statistical correlation between the two, which can be established as follows: (1) Sample preparation and index definition Grease oxidation degree index: Use peroxide value (PV) as a quantitative index (unit: meq / kg) to reflect the content of peroxide in grease (the higher the value, the more severe the oxidation). The normal skin surface grease PV value is usually <5 meq / kg, and ≥8 meq / kg is considered abnormal oxidation.

[0056] Sample size: 2000+ samples (covering different oxidation states), including healthy skin (PV < 5), mild oxidation (5 ≤ PV < 8), moderate oxidation (8 ≤ PV < 12), severe oxidation (> 12), 500+ samples in each group.

[0057] (2) Two-parameter synchronous detection PH value detection: Use the PH sensor module of the component analysis system (precision ±0.05) to measure the skin surface PH value (normal range 4.5-6.5) in the area where facial sebum secretion is active (such as t-zone), and record the stable value.

[0058] Peroxide value detection: Collect skin surface oil samples by tape stripping method, and determine PV value by laboratory standard titration method (GB / T 5538-2005) as "oxidation degree true value".

[0059] (3) Model construction and training Feature analysis: Through correlation analysis, it is found that the increase of PH value (alkaline) is positively correlated with the oxidation degree of sebum (R=0.78), because alkaline environment can accelerate the activity of lipase and promote the hydrolysis and oxidation of sebum.

[0060] Model selection: Use multiple linear regression model (considering PH value and interaction term), input is PH value (x), output is sebum oxidation degree (y), equation form: y = k1x + k2x 2 + b (k1, k2 are coefficients, b is constant term) For example, the results of a certain model: y = 3.2x² - 25.6x + 52.3 (R²≥0.85).

[0061] Grading mapping: Convert the PV value output by the model into sebum secretion grade (1-5 grade), corresponding relationship: 1st grade (PV < 5) → normal; 2nd grade (PV = 5-8) → mild oxidation; 3rd grade (PV = 8-12) → moderate oxidation; 4th grade (PV = 12-15) → severe oxidation; 5th grade (PV > 15) → severe oxidation.

[0062] (4) Model verification and iteration Cross-validation: Use 5-fold cross-validation method to ensure that the average error of the model on different subsets is ≤0.5 grade.

[0063] Environmental factor correction: Introduce temperature and humidity variables as adjustment items (such as in high temperature and high humidity environment, the influence coefficient of PH value on oxidation degree increases by 15%), to improve the applicability of the model in dynamic environment.

[0064] Through the above method, the two models can realize accurate conversion from sensor detection value to physiological indicators, provide quantitative basis for skin quality analysis, and the construction process based on large samples ensures the universality of the model in different populations.

[0065] In the embodiment of the application, the key parameters to be monitored can include: (1) Detection range: water content 0-80%, oil content 0-50 μg / cm².

[0066] (2) Response time: ≤2 seconds / point, data repeatability error ≤3%.

[0067] In the embodiment of the application, the environment monitoring device for acquiring environmental parameters can be implemented in the following manner: Real-time monitoring of temperature and humidity by using built-in temperature and humidity sensors; Real-time acquisition of ultraviolet intensity, PM2.5 and / or wind data through API interface connection to the national meteorological data platform. Specifically, the ambient environmental data of the device can be monitored in real time by using the built-in temperature and humidity sensor (SHT35, accuracy ±0.3℃ / ±2% RH). The ultraviolet intensity (UVI 0-11+), PM2.5, wind and other data of the user's area can be acquired through the API interface connection to the national meteorological data platform, and the update frequency is 1 hour / time. The sensor sampling frequency can be 1 minute / time, and the data storage capacity can be ≥3 months. The meteorological data delay can be ≤15 minutes.

[0068] The core process can be: (1) Data fusion: integrate the measured temperature and humidity of the sensor with the meteorological data to establish an environmental influence coefficient model. (2) Early warning mechanism: when the ultraviolet intensity is ≥7, automatically trigger the "strengthen sunscreen" prompt, and synchronously feedback to the formula design system.

[0069] Fusion of environmental data can provide environmental dimension reference for formula design.

[0070] Specifically, multivariate linear regression combined with random forest model can be used to analyze the correlation between environmental factors such as temperature, humidity, ultraviolet intensity and skin quality changes. Time series analysis is introduced to capture the lagging effect of dynamic changes of environmental factors on skin quality. Training data: environmental data and corresponding skin quality change data in different regions and different seasons, with a total sample size of 100,000+.

[0071] The processing flow can include: data fusion: spatiotemporal alignment of sensor data and meteorological data, construction of a unified data set. Correlation analysis: calculate the importance weight of each environmental factor on skin quality through random forest. Influence coefficient calculation: based on the multiple linear regression model, get the skin quality influence coefficient under different environmental factors (such as the oil secretion aggravation coefficient under high temperature and high humidity environment. Optimization mechanism: according to the actual environmental data of the region where the user is located, regularly update the model parameters, improve the prediction accuracy of the influence of the specific regional environment.

[0072] The mapping relationship model between the basic data and the formula is obtained by training according to the following method: Obtain training data, the training data comprising: basic data, corresponding formula and effect feedback data; Adjust the skin quality index according to the environmental parameter to obtain the adjusted skin quality index; Determine the skin quality type based on the adjusted skin quality index and the skin component; Use the determined skin quality type and the corresponding formula to train a deep reinforcement learning model; in the training process, use the effect feedback data to update the reward function of the model.

[0073] The mapping relationship model between the basic data and the formula can adopt a deep reinforcement learning model, and the training data covers 5000+ cosmetic raw material characteristics (such as skin feel of moisturizing agent, applicable concentration of preservative), 300,000+ skin quality-formula matching cases. Load knowledge graph, associate raw material efficacy (such as ceramide -> repair barrier), contraindicated ingredients (such as alcohol -> sensitive skin) and environmental adaptation rules (such as high humidity -> light lotion).

[0074] The implementation of the knowledge graph in the cosmetic raw material association system is to convert discrete knowledge such as raw material efficacy, contraindicated ingredients and environmental adaptation into a calculable association network through the four-step architecture of entity definition-relation modeling-rule embedding-reasoning engine. The following is the specific implementation method: Knowledge graph design: define core entities and relationships.

[0075] First, build a bottom-layer data model to clearly define "who and who have what relationship", the core entities and relationships are as follows: The following is an example of a triple (entity-relation-entity): ceramide-[has efficacy]->repair barrier; alcohol-[not suitable for]->sensitive skin; high humidity-[recommended]->light lotion.

[0076] In the embodiments of the present application, data collection and knowledge extraction can be performed in the following method: constructing a structured knowledge base. Extracting entities, relationships and attributes from multi-source data to form the "data skeleton" of the knowledge graph: 1. Data sources can include: Authority database: INCI raw material database (ingredient standard name), CosIng (EU cosmetic raw material database), FDA cosmetic raw material safety manual (prohibited ingredient list).

[0077] Literature and standards: PubMed (raw material efficacy research literature), "Cosmetic Safety Technical Specifications" (raw material use restrictions), industry white paper (environmental impact on skin quality research).

[0078] Enterprise data: raw material measured data (such as the stability of a certain ceramide at 30% humidity), customer feedback (such as adverse reaction cases of sensitive skin to alcohol).

[0079] 2. Knowledge extraction method Entity recognition: Rule method: extract INCI name (such as "Ceramide NP") and efficacy terms (such as "antioxidant" and "repair barrier") through regular expressions.

[0080] Machine learning: use BERT model to train entity recognizer to automatically identify "raw material-efficacy" entity pairs (such as extracting "hyaluronic acid" and "moisturizing" from "hyaluronic acid can enhance skin moisturizing") from unstructured text (such as research papers).

[0081] Relationship extraction: Based on templates: define semantic templates (such as "X can Y" -> X-[has efficacy]->Y; "X is not suitable for Y" -> X-[not applicable]->Y) and match relationships from text.

[0082] Remote supervision: use known relationships (such as "alcohol is not suitable for sensitive skin") as seeds to find similar sentence patterns (such as "ethanol is harmful to sensitive skin") in a large amount of text, and automatically expand relationship instances.

[0083] Attribute filling: Extract raw material concentration range (such as "ceramide addition amount 0.1%-1%") and efficacy data (such as "vitamin C whitening efficiency increases by 20% at 1% concentration") from literature as entity attributes.

[0084] 3. Knowledge graph storage and visualization: build an associated network with a graph database Store knowledge in a graph database Neo4j (better at handling multi-hop association queries than relational databases), structure as follows: Nodes: Each entity is a node, with attributes attached (e.g. ingredient node contains "INCI Name = Ceramide NP" "Concentration Range = 0.1%-1%"). Edges: Relationships between nodes are edges, with weights attached (e.g. weight of "Ceramide - Repair Barrier" = 0.9, representing the credibility of this association).

[0085] Show the association network through visualization tools (such as Neo4j Bloom), intuitively present the global relationship of "ingredient - efficacy - skin type - environment", and facilitate manual verification and rule optimization.

[0086] 4. Reasoning engine: realize intelligent association and decision support Based on the relationship of the knowledge graph, realize three core functions through rule reasoning + graph algorithm: 1. Accurate matching of ingredients and efficacy Scenario: Recommend suitable ingredients based on "repair barrier" requirements.

[0087] Implementation: Use graph traversal algorithms (such as depth-first search) to start from the "repair barrier" efficacy node, find all ingredient nodes with "HAS_EFFECT" relationship, sort them by relationship weight (credibility), and output the top 5 ingredients (such as ceramide, cholesterol, fatty acids).

[0088] 2. Automatic avoidance of forbidden ingredients Scenario: Exclude forbidden ingredients for sensitive skin.

[0089] Implementation: Define rules: If the user's skin type is sensitive and the ingredient node has a "not suitable for" relationship with the sensitive skin node, mark it as forbidden.

[0090] Reasoning process: When generating a formula, call the rule engine to check the ingredient list and automatically exclude nodes such as "alcohol" and "fragrance" (confirmed through multi-hop query: sensitive skin - [forbidden] -> alcohol).

[0091] 3. Dynamic adaptation of environment-dosage form Scenario: Recommend suitable dosage form in high humidity environment.

[0092] Implementation: Build environment-skin impact chain: high humidity - [leads to] - increased skin oil secretion - [needs] - oil control / light dosage form.

[0093] Use path query algorithm to find the "high humidity - recommend - light lotion" path, and associate the characteristic ingredients of the lotion (such as containing silica to adjust skin feel), to ensure that the formula adapts to the environment.

[0094] 4. Knowledge Update and Iteration: Keep the Atlas Timely Automatic Update: Connect with authoritative database APIs (such as CosIng update notifications), when raw material safety information changes (such as a component is newly added as sensitive skin taboo), automatically update the corresponding node relationship (such as add "not applicable" edge).

[0095] Manual Verification: Newly added relationships are reviewed by cosmetic R&D experts for credibility (such as new efficacy research in literature), and the weight is updated after verification (such as from 0.6 to 0.8).

[0096] Feedback Optimization: Collect formulation usage effect data (such as a raw material's effect decreases in low humidity), and correct the "raw material - environment" adaptation relationship in the opposite direction (such as reduce the recommended weight of this raw material in low humidity environment).

[0097] 5. Formulation design can adopt the following methods (1) Core Function: Generate personalized skincare formulations based on multi-dimensional data, and dynamically adjust the formulation according to skin condition and environmental changes.

[0098] (2) Technical Architecture: The core algorithm can use a deep reinforcement learning model, aiming to improve skin condition, and continuously optimize the formulation strategy through interaction with the environment (skin and environmental data). Combined with the knowledge graph, the characteristics, efficacy, and taboo of raw materials are integrated into the model reasoning process to ensure the safety and effectiveness of the formulation. Training data: contains 5000+ cosmetic raw material characteristics data, 300,000+ skin - formulation matching cases, and corresponding usage feedback data.

[0099] Processing flow: Data input: Receive multi-dimensional data from imaging skin testing, ingredient analysis, and environmental testing.

[0100] State evaluation: Evaluate the current skin condition and environmental impact based on input data, and determine the formulation design goal.

[0101] Formulation generation: Generate an initial formulation through a deep reinforcement learning model, and conduct compliance and applicability checks in combination with the knowledge graph.

[0102] Dynamic adjustment: Continuously monitor skin changes, and automatically adjust the formulation ingredients and proportions when the changes exceed the set threshold.

[0103] Optimization mechanism: Collect user usage feedback, update the model reward function with feedback data, and continuously improve the actual effect of the formulation.

[0104] In an embodiment of the present application, the step of adjusting the skin condition index according to the environmental parameters can include the following steps: A multi-random forest model was used to determine the importance weights of each environmental parameter on skin quality indicators; Based on the importance weights of each environmental parameter, a multiple linear regression model is used to determine the influence coefficient of the comprehensive environment on skin quality indicators. The skin quality index is adjusted based on the influence coefficient to obtain the final skin quality index.

[0105] In this embodiment of the invention, the compliance and suitability check of the initial formula using a knowledge graph of associated cosmetic ingredient characteristic data includes: The entities associated with the knowledge graph include: raw materials, efficacy, skin type, environmental factors, and cosmetic dosage forms; Based on knowledge graphs, the system uses rule-based reasoning and graph algorithms to match raw materials and efficacy, avoid prohibited ingredients, and adapt to different environments and dosage forms.

[0106] In this embodiment of the invention, the step of matching the physical materials corresponding to the components of the final formula into the corresponding component bins 3 (e.g., Figure 2 (as shown), including: The final formula's ingredient efficacy requirements are translated into the characteristic indicators of the physical material; Entities with a matching degree of ≥90% are selected from the raw material database using the cosine similarity algorithm; The GNN model is used to evaluate the compatibility between candidate entity materials and eliminate conflicting combinations. When users manually replace physical materials, a compatibility verification is performed again.

[0107] Among them, the raw material database integrates the INCI standard raw material library (containing 8,000+ ingredients), labeling the efficacy, suitable skin type, safe dosage and environmental compatibility of each ingredient (such as vitamin C → better stability in a light-protected environment).

[0108] The ingredient compartments can be modularly designed, with each compartment corresponding to a single ingredient and an RFID tag built in to record the ingredient batch and shelf life.

[0109] The core process may include: (1) Matching Logic: Based on the functional requirements of ingredients in the AI ​​formula (e.g., "moisturizing + anti-oxidation"), ingredients with a functional matching degree ≥90% (e.g., hyaluronic acid + ergothioneine) are selected from the ingredient database, excluding ingredients that the user is allergic to (based on preset user questionnaires). Allergenic ingredients are excluded through preset selection. (2) Inventory management: Real-time monitoring of ingredient inventory levels. When an ingredient falls below a threshold (e.g., 50 uses), a replenishment reminder is pushed to the user's APP.

[0110] (3) Replacement mechanism: Support users to manually replace ingredients (e.g. replace panthenol with asiaticoside), and the system automatically verifies compatibility (e.g. pH matching, stability). Key parameters can include: ingredient matching accuracy ≥ 95%, and ingredient bin supporting storage of ≥ 30 ingredients.

[0111] According to the requirements of the formula and the skin quality of the user, the appropriate raw material ingredients can be accurately matched, and the replacement and compatibility verification of the ingredients can be supported. Specifically, the cosine similarity algorithm is used to calculate the matching degree of the formula requirements and the characteristics of the raw materials, and the rule-based reasoning is used to exclude unsuitable ingredients. The graph neural network (GNN) is introduced to analyze the compatibility between ingredients, and a component correlation graph is constructed. The training data includes INCI standard data of 8000+ raw materials, ingredient efficacy data, suitable skin quality data, and ingredient compatibility experimental data.

[0112] The processing flow can include: requirement analysis: converting the efficacy requirements of the formula (such as moisturizing, antioxidant) into specific ingredient property indicators. Matching calculation: filtering ingredients with a matching degree ≥ 90% from the ingredient database through the cosine similarity algorithm. Compatibility verification: using the GNN model to evaluate the compatibility between candidate ingredients, and excluding ingredient combinations with conflicts. Replacement verification: when the user manually replaces ingredients, re-verify the compatibility and applicability to ensure the effectiveness of the replaced formula. Continuously incorporate new ingredient data and compatibility experimental results to update the ingredient correlation graph and matching algorithm, and improve the matching accuracy.

[0113] In an embodiment of the present application, the material taking unit can adopt a precise syringe pump sampling module, cooperate with a multi-channel switching valve (material: 316L stainless steel), support 0.01-10mL range sampling, and the accuracy is ±0.5%. The integrated weight calibration system (accuracy 0.1mg) can correct the sampling error in real time.

[0114] Specifically, the core process can include: (1) Proportion analysis: converting the mass ratio of the formula (such as A:B=3:1) into the sampling volume (based on the ingredient density parameter).

[0115] (2) Sampling execution: sequentially extracting raw materials from the ingredient bin according to the formula order, and after each sampling, the pipeline is blown by inert gas to avoid cross contamination.

[0116] (3) Calibration feedback: if the actual sampling amount deviates from the theoretical value by ≥ 1%, automatically supplement or empty and re-take.

[0117] Among them, the maximum sampling speed can be 5mL / min, which can support up to 8 ingredients to be sampled at the same time.

[0118] In one embodiment of the present invention, the mixing chamber of the preparation unit can adopt a double-jacket design (temperature control range can be 15-40℃), with a built-in magnetic stirrer (100-1000 rpm) and a high-speed homogenizer (1000-6000 rpm), and can be optionally equipped with a high-pressure microjet module (pressure 50-200 bar). The specific structure and component connection relationship of the preparation mixing chamber are described below.

[0119] Main Structure and Core Components: The mixing chamber, serving as the core unit for material mixing, emulsification, and homogenization, adopts a vertical cylindrical structure (diameter 150-300mm, volume 500-2000mL). It is made of 316L stainless steel (material contact parts) + a heat insulation layer (outer layer). The core consists of five main modules: a double-layer jacketed chamber: an inner layer (material contact) 3mm thick, an outer layer (temperature control) 2mm thick, a 5mm gap between the layers, a media inlet at the top, and a media outlet at the bottom. The chamber temperature is controlled by circulating media (water / heat transfer oil). The magnetic stirring assembly is a bottom-embedded magnetic drive unit (external permanent magnet + internal PTFE stir bar, stir bar diameter 20-50mm), achieving low-shear mixing and preventing material splashing. The high-speed homogenization assembly is a top-suspended rotor-stator structure (rotor diameter 40-80mm, stator with a 200μm aperture screen), connected to a variable frequency motor, enabling high-shear emulsification and dispersing the oil / solid phase into fine particles. Material inlet and outlet interfaces: three raw material inlets (Φ8mm) at the top and one outlet (Φ10mm) at the lower side, all equipped with quick-connect valves to control material inflow / outflow and facilitate cleaning and pipeline replacement; high-pressure microjet interface (optional). The outlet is connected to a high-pressure microjet module (imported pressure sensor + gemstone nozzle, orifice diameter 0.1-0.5mm) via a three-way valve, achieving nanoscale dispersion through ultra-high pressure shearing (50-200bar).

[0120] The component connections are as follows: [Temperature Control System] → Medium Pump → Double Jacket (Upper Inlet) → Jacket Layer → (Lower Outlet) → Back to Temperature Control System → [Raw Material Storage Tank 1] → Metering Pump → Raw Material Inlet 1 → Inside Mixing Chamber [Raw Material Storage Tank 2] → Metering Pump → Raw Material Inlet 2; [High-Speed ​​Homogenizing Motor] → Drive Shaft (Mechanical Seal) → Rotor-Stator Assembly (Top Inside the Chamber); [Magnetic Stirring Driver] (Bottom Outside the Chamber) → Stirrer (Bottom Inside the Chamber); Mixing Chamber → Discharge Valve → Three-Way Valve → [Regular Discharge] → [High-Pressure Micro-Jet Module] → High-Pressure Nozzle → Finished Product Discharge. Key connection details: 1. Double jacket and temperature control system connection - The upper inlet of the jacket is connected to a constant temperature circulator (such as Huber CC-100) through a Φ12mm silica gel pipe, and the lower outlet is returned through a pipe of the same specification, forming a closed loop; a flow sensor (accuracy ±2%) is installed on the pipe to real-time feedback the medium flow rate (1-5L / min), ensuring the temperature control accuracy (±0.5℃). 2. Homogenization assembly and drive system connection - A high-speed homogenization motor (power 500-1500W) is installed on the top flange of the mixing chamber, penetrating the chamber through a transmission shaft, and the shaft and the chamber contact part use double-end mechanical seals (material: silicon carbide + fluorine rubber) to prevent material leakage (pressure resistance ≤0.6MPa); the transmission shaft end connects the rotor, which cooperates with the stator fixed on the inner wall of the chamber, and the rotor speed is adjusted by a frequency converter controller (1000-6000 revolutions / minute). 3. Magnetic stirring and external drive connection: An electromagnetic driver (220V, 50Hz) is installed on the outside of the bottom of the mixing chamber, which drives the stirring rod in the chamber to rotate through an alternating magnetic field (speed 100-1000 revolutions / minute), without mechanical contact, avoiding the risk of bottom leakage; the stirring rod bottom is provided with a polytetrafluoroethylene support point to prevent friction with the chamber bottom during high-speed rotation and generate particles. 4. Optional high-pressure micro-jet module connection: The mixing chamber outlet is switched by a three-way valve: normal mode directly discharges, high-pressure mode connects to the micro-jet module; a pressure sensor (0-300bar) is installed at the inlet of the micro-jet module, and a cooling sleeve (shared with the jacket temperature control system) is provided at the outlet to ensure that the material temperature does not exceed 40℃ after high-pressure treatment. 5. Control system integration: All motors, valves, and sensors are connected through a PLC (such as Siemens S7-1200), and the touch screen displays real-time temperature, speed, and pressure parameters, supporting preset programs (such as "stirring 300 revolutions / 5 minutes → homogenization 3000 revolutions / 3 minutes → micro-jet 100bar / 2 minutes").

[0121] Collaborative workflow 1. The material enters the mixing chamber through the top inlet, and the double jacket is connected to the temperature control medium (such as 35℃ water) to maintain the set temperature; 2. The magnetic stirrer rotates at low speed (300 revolutions / minute) to preliminarily mix the water-soluble ingredients; 3. The high-speed homogenizer starts (3000 revolutions / minute) to shear the oil phase ingredients into 1-10μm particles, achieving emulsification; 4. If higher dispersion (such as nanoscale essence) is required, switch to the high-pressure micro-jet module (150bar) to refine the particles to 100-500nm through the shearing and collision action of the gemstone nozzle; 5. The final material is discharged from the outlet and enters the next link (such as quantitative filling). Through modular connection and automatic control, this system can adapt to the preparation needs of different dosage forms such as facial cleanser, lotion, and essence, balancing mixing efficiency and material stability. 6. Food-grade PTFE is selected to ensure compatibility with cosmetic raw materials (acid-resistant, alkali-resistant, and organic solvent-resistant).

[0122] The working core process of the preparation unit can be: (1) Stepwise preparation: the water-soluble ingredients are first dissolved (stirring 300 rpm / min), then the oil phase ingredients are added and homogenized (3000 rpm / min), and finally the particles are refined by microfluidization (particle size ≤1 μm).

[0123] (2) Time control: the preparation of cleansing water / makeup water is ≤2 minutes, the preparation of emulsion / cream is ≤5 minutes, and the viscosity change is monitored in real time (target value ±50 cP) throughout the process. The key parameters that need to be controlled can include: mixing uniformity: ≥99%, batch-to-batch reproducibility error ≤3%.

[0124] While the preferred embodiments of the application have been described, modifications and alterations can occur to others upon reading the preceding description in light of the accompanying claims. It is intended that the application be construed as including all such modifications and alterations insofar as they come within the scope of the claimed application. It should be understood by those skilled in the art that various modifications and changes can be made to this application without departing from the spirit and scope of the application. Accordingly, it is intended that all such modifications and changes be included within the scope of the application as long as they fall within the scope of the claims and their equivalents.

Claims

1. An apparatus for intelligent skin-matching customized cosmetics, characterized by, The device comprises: a test unit configured to acquire basic data of a current user, the basic data comprising skin texture indicators, skin components, and environmental parameters; a formula design unit configured to determine initial formulas of various types of cosmetics suitable for the current user based on the acquired basic data and a mapping relationship model between the basic data and the formulas, and to perform compliance and applicability checks on the initial formulas by using a knowledge graph associated with cosmetic raw material characteristics data to obtain final formulas; a material matching unit configured to match entity materials corresponding to components of the final formulas to corresponding component warehouses; a component warehouse configured to store entity materials corresponding to components; a material taking unit configured to take out entity materials corresponding to a corresponding proportion from a corresponding component warehouse based on the final formula and deliver the entity materials to a preparation unit; a preparation unit configured to prepare cosmetics customized for the user by using the entity materials.

2. The apparatus for intelligent skin-customized cosmetic product of claim 1, wherein, The device further comprises a quantitative discharging unit and a cleaning unit; the quantitative discharging unit is configured to discharge the prepared product by driving a high-pressure gas flow; and the cleaning unit is configured to clean and sterilize the material taking unit, the preparation unit, and the quantitative discharging unit.

3. The apparatus for intelligent skin-customized cosmetic product of claim 1, wherein, The test unit comprises a non-contact skin imaging device, a contact skin component testing device, and an environmental monitoring device; the non-contact skin imaging device is configured to acquire skin texture indicators; the contact skin component testing device is configured to acquire skin components; and the environmental monitoring device is configured to acquire environmental parameters.

4. The apparatus for intelligent skin-matching customized cosmetic according to claim 3, wherein, The non-contact skin imaging device is configured to acquire skin texture indicators in the following manner: acquire infrared response images and multispectral images of a face; generate a face texture map based on the infrared response images and the multispectral images; obtain skin texture indicators based on the face texture map by using a deep learning model.

5. The apparatus for intelligent skin-matching customized cosmetic according to claim 3, wherein, The contact skin component testing device is configured to acquire skin components in the following manner: detect the conductivity and the pH value of a skin surface; obtain the water content of the skin based on the detected conductivity by using a pre-constructed mapping relationship between the conductivity and the water content; obtain the oxidation degree of oil on the skin based on the detected pH value by using a pre-constructed correlation model between the pH value and the oxidation degree of oil, and determine the oil secretion level based on the oxidation degree of oil.

6. The apparatus for intelligent skin-matching customized cosmetics according to claim 3, wherein, The environmental monitoring device is configured to acquire environmental parameters in the following manner: monitor the temperature and the humidity in real time by using built-in temperature and humidity sensors; obtain ultraviolet intensity, PM2.5, and / or wind force data in real time by connecting to a national meteorological data platform through an API interface.

7. The apparatus of claim 1, wherein the skin analysis module comprises a skin color analysis module. The mapping relationship model between the basic data and the formulas is obtained in the following manner: acquire training data, the training data comprising basic data, corresponding formulas, and effect feedback data; adjust the skin texture indicators based on the environmental parameters to obtain adjusted skin texture indicators; determine a skin texture type based on the adjusted skin texture indicators and the skin components; and A deep reinforcement learning model is trained by using the determined skin type and corresponding formula; during the training process, the effect feedback data is used to update the reward function of the model.

8. The apparatus of claim 1, wherein the skin analysis module comprises a skin color analysis module. The adjusting the skin quality index according to the environmental parameters comprises: An importance weight of the influence of each environmental parameter on the skin quality index is determined by using a multi-random forest model; Based on the importance weight of each environmental parameter, a multivariate linear regression model is used to determine the influence coefficient of the comprehensive environment on the skin quality index; The skin quality index is adjusted based on the influence coefficient to obtain the final skin quality index.

9. The apparatus of claim 1, wherein the skin analysis module comprises a skin color analysis module. The knowledge graph associated with the cosmetic raw material characteristic data is used to perform compliance and applicability checking on the initial formula, comprising: The entities associated with the knowledge graph include: raw materials, effects, skin quality, environmental factors and cosmetic dosage forms; Based on the knowledge graph, raw materials and effects are matched by rule reasoning and graph algorithms to avoid banned ingredients and adapt to the environment and dosage forms.

10. The apparatus of claim 1, wherein the skin analysis module comprises a skin color analysis module. The entity materials corresponding to the components of the final formula are matched to the corresponding component bins, comprising: The component effect requirements of the final formula are converted into characteristic indexes of entity materials; Entity materials with a matching degree ≥90% are screened from the raw material database by using a cosine similarity algorithm; The compatibility between candidate entity materials is evaluated by using a GNN model to exclude combinations with conflicts; When the user manually replaces the entity materials, compatibility verification is performed again.

Images