Cosmetic evaluation method, device, equipment and medium

Abstract

The invention discloses a cosmetic evaluation method, device and equipment and a medium, and relates to the technical field of cosmetic evaluation, and the cosmetic evaluation method comprises the following steps: S1, collecting physical optical fingerprints; s2, carrying out short-term sensory and multi-modal acquisition; s3, time sequence texture and microstructure extraction; s4, accelerated disturbance and response fitting; s5, interpretable coupling modeling is carried out; s6, performing grouping optimization and evaluation output; according to the method, the output of the interpretable coupling model and the layered consumer preference are combined for multi-objective optimization, so that the evaluation result not only represents general performance, but also can directly reflect the preference and weight difference of different user groups to the performance dimension. And for a certain target group, the model can point out the most effective improvement direction and influence factor, so that the closed-loop time from problem identification to formula iteration is shortened. Compared with a single total score, the mechanism enables the evaluation result to be more operable and economic rationality in commercialized positioning, iterative priority setting and small-batch trial-production.

Description

Technical Field

[0001] This invention relates to the field of cosmetic evaluation technology, specifically to a method, apparatus, equipment, and medium for evaluating cosmetics. Background Technology

[0002] Performance evaluation of cosmetics is a core component of research and development, quality control, and commercial positioning, directly determining the product's market suitability and user experience. Currently, cosmetic evaluation methods in the industry still suffer from numerous key technological shortcomings, making it difficult to meet the dual requirements of refined research and development and segmented consumer demands.

[0003] Current technologies for measuring the physical and optical properties of cosmetics, such as reflectance spectrum, viscosity, and particle size distribution, often involve decentralized and independent measurements. These measurements fail to incorporate individual subject information to construct a unified set of physical and optical characteristics. Furthermore, the data for each indicator lacks standardized baseline calibration, numerical normalization, and dimensionality optimization. This results in a lack of cross-comparison between test data from different testing scenarios and batches. It also makes it difficult to effectively mitigate the interference of individual skin differences on evaluation results, thus failing to establish a stable reference system for the fundamental characteristics of the product.

[0004] In response to this, this application proposes a method, apparatus, equipment, and medium for evaluating cosmetics to solve the aforementioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a method, apparatus, equipment and medium for evaluating cosmetics, in order to solve the problems in the prior art where test data from different testing scenarios and batches lack horizontal comparability, and it is also difficult to effectively avoid the interference of individual skin differences on the evaluation results, and it is impossible to form a stable reference system for the basic characteristics of products.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for evaluating cosmetics, comprising:

[0008] S1. Obtain the cosmetic sample and the subject's baseline information (including age, skin type, recent skin care history, etc.), and perform initial physical and optical characterization on the cosmetic sample under standardized light and temperature and humidity conditions (including but not limited to reflectance spectrum, color, gloss, viscosity and particle size distribution) to obtain the physical optical fingerprint (a structured multidimensional feature vector containing standardized values ​​of spectrophotometric reflectance spectrum, gloss / brightness curve, viscosity curve, particle size distribution, etc.).

[0009] S2. Applying samples to the skin and simultaneously preprocessing signals: The physical optical fingerprint is bound to a preset short scale, and samples are applied to the local skin of the subject in a uniform time sequence while simultaneously acquiring high frame rate video, micro-audio, and subjective short scale scores. The acquired time-series raw signals are then synchronized, denoised, baseline removed, and standardized to obtain a short-term sensory response time series (a time-aligned multi-channel time-series dataset).

[0010] S3. Spatiotemporal alignment and texture / optical feature extraction: The short-term sensory response time series is spatially and temporally registered and aligned with the macro skin images acquired in the same time window. After alignment, spatiotemporal texture and optical evolution features are extracted based on operators of optical flow, local binary mode (LBP), and short-time Fourier transform (STFT) to generate a temporal texture and optical change description (a temporal feature matrix or several aggregated statistics describing the behavior of materials on the skin surface).

[0011] S4. Accelerated test and dynamic fitting under controllable perturbation: The temporal texture and optical change description are superimposed with the preset controllable perturbation parameters (including temperature level, relative humidity, number of frictions, simulated sweating intensity) according to the test design to execute a short-term acceleration scenario, and the temporal response under each scenario is fitted by a function to calculate and output the acceleration response curve and its dynamic parameters (such as decay rate λ, polynomial coefficients, etc.).

[0012] S5. Coupled Causal Screening and Interpretable Regression Modeling: The accelerated response curve and the synchronously collected non-invasive skin short-spectrum indicators (such as local pH, epidermal temperature, color / redness, sebum content, etc.) are input into the causal discovery or variable screening process to screen candidate influence paths, and an interpretable coupled model is established by fitting the coupling relationship between sensory changes and microphysiological responses with sparse regularized interpretable regression or Bayesian regression.

[0013] S6. Cluster Preference Mapping and Multi-Objective Optimization Output Evaluation: The output of the interpretable coupled model is mapped to a hierarchical consumer preference database (preference weight vectors for clusters based on skin type / age / culture). The performance score for each cluster is calculated through clustering and multi-objective optimization (weighted sum or Pareto analysis). The performance score, feature importance, and targeted improvement suggestions are summarized in an interpretable report to output the clustered performance score and the final evaluation report.

[0014] Furthermore, the cosmetic sample and the subject's baseline information are collected, and the cosmetic sample is subjected to initial physical and optical characterization under standardized light and temperature / humidity conditions to obtain a physical-optical fingerprint, including:

[0015] The reflectance spectrum R(λ) and viscosity curve η(·) were obtained by spectrophotometry (wavelength range 400–700 nm) and rotational viscometer. The particle size distribution was measured by dynamic light scattering to obtain d50 for initial physical and optical characterization to obtain physical and optical fingerprints.

[0016] The physical optical fingerprint is obtained by principal component dimensionality reduction of the above features after baseline correction and normalization.

[0017] Furthermore, the physical optical fingerprint is bound to a preset short scale, and samples are applied to the subject's local skin in a uniform time sequence while simultaneously acquiring high frame rate video, micro-audio, and subjective short scale scores. The acquired raw time-series signals are then synchronized, denoised, baseline-removed, and standardized to obtain a short-term sensory response time series, including:

[0018] Multimodal data acquisition at local skin sites of subjects was performed using high frame rate video acquisition with a frame rate of no less than 120 fps and an audio sampling rate of no less than 16 kHz. Subjective short scales were recorded at preset sampling points t={0,30,60,120,300,600 s}. The acquired raw time series data were uniformly timestamped, bandpass filtered for noise reduction and zero drift, and baseline removed to obtain short-term sensory response time series.

[0019] Further, the step of spatial-temporal registration and alignment of the short-term sensory response time series with macro skin images acquired within the same time window, followed by extraction of spatiotemporal texture and optical evolution features based on operators of optical flow, local binary mode (LBP), and short-time Fourier transform (STFT) to generate a temporal texture and optical change description from the short-term sensory response time series and macro skin images, includes:

[0020] First, the video frames are registered in time and space. Then, the surface displacement field u(x,y,t) is extracted based on the optical flow method. The local binary mode (LBP) histogram and local contrast are calculated to generate temporal texture and optical change description.

[0021] The temporal texture and optical variations are described as a matrix representation of LBP differences, brightness variation curves, and particle visibility indices aggregated by time windows.

[0022] Furthermore, the process of superimposing the temporal texture and optical change description with preset controllable perturbation parameters (including temperature level, relative humidity, number of friction cycles, and simulated sweating intensity) according to the experimental design to execute a short-term acceleration scenario, and performing function fitting on the temporal response under each scenario to calculate and output the acceleration response curve and its dynamic parameters, including:

[0023] At least three temperature ranges and three friction cycles were combined, and under each disturbance scenario, the time-series response was fitted with exponential decay or polynomial fitting using nonlinear least squares method to obtain the acceleration response curve and its dynamic parameters (including but not limited to decay rate λ and goodness of fit R). 2 The perturbation is applied on the timeline and recorded with event timestamps to ensure precise alignment with the time-series data;

[0024] The controllable disturbance parameters include: simulation by superimposing three increments of temperature (e.g., 25°C, 32°C, 38°C) with a preset number of friction cycles (e.g., 0, 10, 50 times), with each disturbance level converted to equivalent daily use on the time axis using an acceleration coefficient α;

[0025] The acceleration response curve is obtained by fitting a polynomial or exponential decay function.

[0026] Furthermore, the accelerated response curve and simultaneously acquired non-invasive short-spectrum skin indicators (such as local pH, epidermal temperature, chroma / redness, sebum content, etc.) are input into a causal discovery or variable screening process to screen candidate influencing paths. An interpretable coupling model is then established by fitting the coupling relationship between sensory changes and microphysiological responses using sparse regularized interpretable regression or Bayesian regression. This includes:

[0027] First, a constraint-based causal discovery algorithm (such as PC algorithm) is applied to the candidate variable set to filter out candidate causal edges. Then, the interpretable coupling model H is obtained by fitting the generalized linear regression (LASSO-GLM) with L1 sparse regularization or Bayesian regression. The regularization parameter is selected through cross-validation to output the effect coefficient, significance test result and confidence interval of each input feature in the interpretable coupling model.

[0028] Furthermore, the output of the interpretable coupling model is mapped to a hierarchical consumer preference database (preference weight vectors grouped by skin type / age / culture preference). Performance scores for each group are calculated through clustering and multi-objective optimization (weighted sum or Pareto analysis). The performance scores, feature importance, and targeted improvement suggestions are summarized in an interpretable report to output a grouped performance score and a final evaluation report, including:

[0029] The number of clusters in the hierarchical consumer preference database is determined based on the silhouette coefficient, and k-means clustering is used to obtain the groups. For each group, the immediate sensory score, durability index and skin compatibility are used as optimization objectives. The weighted sum method or Pareto front analysis is used to perform multi-objective optimization to calculate the grouping performance score, and the actionable improvement suggestions based on model H and the corresponding uncertainty assessment are output in priority order.

[0030] Another aspect of the present invention is to provide an evaluation device for cosmetics, comprising:

[0031] The physical optical fingerprint acquisition module is used to acquire cosmetic samples and subject baseline information, and to perform initial physical and optical characterization of the cosmetic samples under standardized light and temperature and humidity conditions to obtain physical optical fingerprints;

[0032] The short-term sensory and multimodal acquisition module is used to sample and simultaneously acquire high frame rate video, micro-audio, and subjective short scale scores to perform synchronous time synchronization, noise reduction, baseline removal, and standardization on the acquired raw time signals to obtain short-term sensory response time series.

[0033] The temporal texture and microstructure extraction module is used to perform spatial-temporal registration and alignment between the short-term sensory response time series and the macro skin images acquired in the same time window. After alignment, the spatiotemporal texture and optical evolution features are extracted based on operators of optical flow, local binary mode, and short-time Fourier transform to generate a temporal texture and optical change description.

[0034] The acceleration perturbation and response fitting module is used to superimpose the temporal texture and optical change description with the preset controllable perturbation parameters according to the experimental design to execute short-term acceleration scenarios, and to perform function fitting on the temporal response under each scenario to calculate and output the acceleration response curve and its dynamic parameters.

[0035] An interpretable coupling modeling module is used to input the accelerated response curve and the synchronously acquired non-invasive skin short-spectrum index into the causal discovery or variable screening process to screen candidate influence paths, and to fit the coupling relationship between sensory changes and microphysiological responses to establish an interpretable coupling model.

[0036] The cluster optimization and evaluation output module is used to map the output of the interpretable coupled model to a hierarchical consumer preference database, calculate the performance score of each cluster through clustering and multi-objective optimization, and summarize the performance score, feature importance and targeted improvement suggestions in the form of an interpretable report to output the clustered performance score and the final evaluation report.

[0037] An electronic device, the electronic device comprising:

[0038] At least one processor; and a memory communicatively connected to said at least one processor; wherein:

[0039] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described above.

[0040] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0041] Compared with existing technologies, the present invention provides a method, apparatus, device, and medium for evaluating cosmetics. This application synchronously couples real-time subjective feelings, optical / acoustic temporal signals, and microstructural images along the time axis, and extracts dynamic features under perturbation conditions. This enables the establishment of a clear causal or correlation path between macroscopic surface performance and microscopic morphological evolution. This temporal coupling allows the evaluation to move beyond static endpoints or single indicators, enabling the identification of specific physical processes (such as particle aggregation, film rupture, and oil absorption / separation processes) that lead to performance degradation or improvement. This elevates "performance evaluation" to "mechanism identification."

[0042] By combining the output of an interpretable coupled model with hierarchical consumer preferences for multi-objective optimization, the evaluation results not only characterize general performance but also directly reflect the differences in preferences and weights of different user groups for performance dimensions. This segmented scoring system provides a clear decision-making path for R&D: for a specific target segment, the model can point out the most effective improvement direction and influencing factors, thereby shortening the closed-loop time from problem identification to formulation iteration. Compared to a single overall score, this mechanism makes the evaluation results more operable and economically reasonable in terms of commercial positioning, iteration priority setting, and small-batch trial production. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0044] Figure 1 A flowchart illustrating a cosmetic evaluation method provided in an embodiment of the present invention;

[0045] Figure 2 This is a block diagram of a cosmetic evaluation device provided in an embodiment of the present invention. Detailed Implementation

[0046] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0047] As attached Figure 1 As shown:

[0048] Example 1:

[0049] A method for evaluating cosmetics, comprising:

[0050] S1. Obtain cosmetic samples and subject baseline information, and perform initial physical and optical characterization of the cosmetic samples under standardized light and temperature and humidity conditions to obtain physical optical fingerprints;

[0051] In step S1, the baseline information of the subjects includes age, skin type, recent skin care history, etc.

[0052] Initial physical and optical characterization includes reflectance spectrum, colorimetry, gloss, viscosity, and particle size distribution;

[0053] Physical optical fingerprints are structured multidimensional feature vectors containing standardized values ​​such as spectrophotometric reflectance spectrum, gloss / brightness curve, viscosity curve, and particle size distribution.

[0054] Furthermore, the reflectance spectrum R(λ) and viscosity curve η(·) were obtained by spectrophotometry (wavelength range 400–700 nm) and rotational viscometer, and the particle size distribution was measured by dynamic light scattering to obtain d50 for initial physical and optical characterization to obtain physical-optical fingerprint.

[0055] The physical optical fingerprint is obtained by dimensionality reduction of the above features after baseline correction and normalization according to principal components.

[0056] S2. Applying samples to the skin and simultaneously preprocessing signals: The physical optical fingerprint is bound to a preset short scale. Samples are applied to the local skin of the subject in a uniform time sequence, and high frame rate video, micro-audio and subjective short scale scores are collected at the same time. The collected time sequence raw signals are then synchronized, denoised, baseline removed and standardized to obtain a short-term sensory response time sequence.

[0057] In step S2, multimodal acquisition at the local skin of the subject is carried out using high frame rate video acquisition with a frame rate of not less than 120 fps and audio sampling rate of not less than 16 kHz. Subjective short scale is recorded according to preset sampling points t={0,30,60,120,300,600s}. The acquired raw time series data is uniformly timestamped, bandpass filtered for noise reduction and zero drift, and baseline removal is performed to obtain short-term sensory response time series.

[0058] S3. Spatiotemporal alignment and texture / optical feature extraction: The short-term sensory response time series is spatially and temporally registered and aligned with the macro skin images acquired in the same time window. After alignment, spatiotemporal texture and optical evolution features are extracted based on operators of optical flow, local binary mode (LBP), and short-time Fourier transform (STFT) to generate a temporal texture and optical change description.

[0059] In step S3, the video frames are first registered in time and space, and then the surface displacement field u(x,y,t) is extracted based on the optical flow method. The local binary mode (LBP) histogram and local contrast are calculated to generate temporal texture and optical change description.

[0060] The temporal texture and optical changes are described as a matrix representation of LBP differences, brightness change curves, and particle visibility indices aggregated by time windows;

[0061] S4. Accelerated test and dynamic fitting under controllable perturbation: The temporal texture and optical change description are superimposed with the preset controllable perturbation parameters (including temperature level, relative humidity, number of frictions, simulated sweating intensity) according to the test design to execute a short-term acceleration scenario, and the temporal response under each scenario is fitted by a function to calculate and output the acceleration response curve and its dynamic parameters (such as decay rate λ, polynomial coefficients, etc.).

[0062] In step S4, at least three temperature levels and three friction cycles are combined, and under each disturbance scenario, the time series response is subjected to exponential decay fitting or polynomial fitting using the nonlinear least squares method to obtain the accelerated response curve and its dynamic parameters (including but not limited to decay rate λ and goodness of fit R2). The application of the disturbance on the time axis is recorded with event timestamps to accurately align with the time series data.

[0063] The controllable disturbance parameters include: simulation by superimposing three increments of temperature (e.g., 25°C, 32°C, 38°C) with a preset number of friction cycles (e.g., 0, 10, 50 times), with each disturbance level converted to equivalent daily use on the time axis using an acceleration coefficient α;

[0064] The acceleration response curve is obtained by fitting a polynomial or exponential decay function;

[0065] S5. Coupled Causal Screening and Interpretable Regression Modeling: The accelerated response curve and the synchronously collected non-invasive skin short-spectrum indicators (such as local pH, epidermal temperature, color / redness, sebum content, etc.) are input into the causal discovery or variable screening process to screen candidate influence paths, and an interpretable coupled model is established by fitting the coupling relationship between sensory changes and microphysiological responses with sparse regularized interpretable regression or Bayesian regression.

[0066] In step S5, firstly, a constraint-based causal discovery algorithm (such as PC algorithm) is applied to the candidate variable set to filter out candidate causal edges. Then, the interpretable coupling model is obtained by fitting the generalized linear regression (LASSO-GLM) with L1 sparse regularization or Bayesian regression. The regularization parameter is selected through cross-validation to output the effect coefficient, significance test result and confidence interval of each input feature in the interpretable coupling model.

[0067] S6. Cluster Preference Mapping and Multi-Objective Optimization Output Evaluation: The output of the interpretable coupled model is mapped to a hierarchical consumer preference database (preference weight vectors for clusters based on skin type / age / culture). The performance score for each cluster is calculated through clustering and multi-objective optimization (weighted sum or Pareto analysis). The performance score, feature importance, and targeted improvement suggestions are summarized in an interpretable report to output the clustered performance score and the final evaluation report.

[0068] In step S6, the number of clusters in the hierarchical consumer preference database I is determined based on the silhouette coefficient, and k-means clustering is used to obtain the groups. For each group, the immediate sensory score, durability index and skin compatibility are used as optimization objectives. The weighted sum method or Pareto front analysis is used to perform multi-objective optimization to calculate the grouping performance score, and the actionable improvement suggestions based on model H and the corresponding uncertainty assessment are output in priority order.

[0069] As shown above, this method synchronously couples real-time subjective perception, optical / acoustic temporal signals, and microstructure images along the time axis, and extracts dynamic features under perturbed conditions. This enables the establishment of a clear causal or correlation path between macroscopic surface performance and microscopic morphological evolution. This temporal coupling allows evaluation to move beyond static endpoints or single indicators, enabling the identification of specific physical processes (such as particle aggregation, film fracture, and oil absorption / separation processes) that lead to performance degradation or improvement. Thus, "performance evaluation" is elevated to "mechanism identification."

[0070] Example 2:

[0071] Test parameters:

[0072] Number of participants: N=90 (divided into 3 groups: Example 1 Group, Example 2 Group, and Comparative Group, n=30 in each group), mixed gender, aged 20–55 years, with approximately balanced proportions within each group according to skin type (dry / combination / oily) (approximately 10 participants per skin type per group). Baseline information for participants included: age, skin type, recent skincare history (whether acid / hormonal products were used in the past week), and history of skin sensitivity.

[0073] Standardized environment: room temperature 22±1°C, relative humidity 45±5%, standard D65 light source illumination, neutral gray background.

[0074] Instruments and data acquisition parameters:

[0075] Spectroreflectance spectrometer: wavelength range 400–700 nm, resolution 1 nm, obtains reflectance spectrum R(λ).

[0076] Viscometer (rotary): measures the viscosity curve η(γ˙) at shear rate and gives the constant shear viscosity value (mPa·s).

[0077] High frame rate camera: ≥120 fps, 1080p resolution, for dynamic recording of surface texture and gloss.

[0078] Micro-audio acquisition: sampling rate 16 kHz, used to record the triboacoustic characteristics during application.

[0079] Macro imaging equipment: magnification 10–20×, resolution ≥5 μm / pixel, used for imaging the microstructure of the skin surface.

[0080] Non-invasive short-spectrum skin indicators: portable pH patch (0.1 pH resolution), skin temperature sensor (±0.1°C), and colorimeter (Lab*) to record redness / infrared optical changes.

[0081] Data storage: timestamp synchronization (all devices synchronize to a unified clock), all time-series data are aligned in seconds.

[0082] The specific application scenario of this embodiment is: moisturizing microemulsion cream (goal: to improve immediate feel and short-term longevity, while taking into account skin compatibility).

[0083] Formulation overview (for differentiation of samples only, not as a limitation of the claims)

[0084] Sample A: Water / oil microemulsion system containing glycerin (5%), sodium hyaluronate (0.1%), low viscosity polymerization aid, and nano-dispersed silicate (oil absorption / gloss balance) – used to improve application smoothness and long-lasting water film.

[0085] Comparative example (commercial reference): Similar moisturizing creams but without nano-dispersed silicates, and with 0.05% hyaluronic acid content.

[0086] Step 1.1 (Obtaining the physical optical fingerprint):

[0087] Experiment: The physical optical fingerprint of the sample was measured by spectroscopic reflectance R(λ) and viscosity η(100s) under standard conditions. -1 (Shear rate 100 s) -1 And measure the particle / droplet size distribution (dynamic light scattering, give the average particle size d50).

[0088] The recorded data:

[0089] The reflectance spectrum at 550 nm has a reflectance R(550) = 0.32;

[0090] η(100 s -1 =1200 mPa·s;

[0091] The average particle size d50 = 0.25 μm.

[0092] These multimodal data are merged into a vector = [R(λ_400…700), η(·), d50] (compressed representation of principal component eigenvectors A_pc).

[0093] Step 1.2 (Obtaining the short-term sensory response time series):

[0094] On-site: On the back of the subject's hand (fixed area 4 cm) 2 Apply 0.05 g and the practitioner applies it three times using the standard technique.

[0095] Subjective short scale: immediate feel (0–10), application smoothness (0–10), instant gloss (0–10), recorded at t=0 s (immediately after application).

[0096] High frame rate video (120 fps) and micro-audio (16 kHz) were simultaneously acquired from t = 0 to t = 600 s (10 minutes).

[0097] The time series obtained are: subjective rating S(t) with sampling points t = {0, 30, 60, 120, 300, 600 s}; video / audio is processed by short time Fourier transform (STFT) to extract spectral energy E_audio(t,f) and optical instantaneous brightness I_video(t) (the key channel is the average brightness of the green channel).

[0098] Step 1.3 (Obtaining temporal texture and optical variation description):

[0099] Macro images were acquired synchronously at t = 0, 60, and 300 s. The surface displacement field u(x,y,t) was calculated using the optical flow method, and the local binary pattern (LBP) histogram H_LBP(t) was extracted.

[0100] Temporal characteristics include:

[0101] ΔLBP(t) = H_LBP(t) - H_LBP(0), ΔI(t) = I_video(t) - I_video(0), and particle visibility index P_vis(t) (based on image contrast) are used to form the D(t) matrix.

[0102] Step 1.4 (Obtaining the accelerated response curve):

[0103] Controllable disturbance parameters: three temperature settings: 25°C (normal temperature), 32°C (close to facial temperature), and 38°C (sweat or high temperature conditions); friction count settings: 0, 10, and 50 times (achieved using a standard mechanical friction device); test duration for each setting is 10 minutes.

[0104] For each scenario, record I_video(t) and S(t), and fit the data using an exponential decay model:

[0105] I(t) = I0 e−λt + CI(t)

[0106] The least squares estimate λ (decay rate) to generate F = {λ(situation)}.

[0107] Fitting results (Example 2): Under the condition of 32°C and 10 friction cycles, λ = 0.0035 s was obtained. -1 ,half life ≈ 198 s (approximately 3.3 minutes).

[0108] Step 1.5 (Establishing a sensory-physiological coupling model)

[0109] Non-invasive parameters G (local pH, epidermal temperature, and Lab* a* changes representing redness) were collected. G(t) was obtained under the same conditions.

[0110] Causal discovery: The PC algorithm (oriented) is used to screen for possible causal pathways in the candidate variable set {S(t) features, ΔI(t), ΔLBP(t), pH(t), Temp(t), a*(t)}.

[0111] Fitting model H: A generalized linear model with L1 regularization (LASSO-GLM) is used.

[0112] Sdur=β0+β1ΔIt_agg+β2ΔLBPt_agg+β3ΔpH+β4Temp+ε

[0113] Where Δ·{t_agg} represents the mean / slope aggregation feature over the key time window (0–300 s). The L1 regularization term λ{L1} is selected using cross-validation to ensure interpretability.

[0114] Step 1.6 (Clustered Scoring and Evaluation Output)

[0115] Construct a hierarchical consumer preference database: Based on 500 historical rating vectors (simulated data or real historical database), cluster skin type / age (k-means, k is determined by the silhouette coefficient to be 3 groups: young oily group, middle-aged combination group, and mature dry group).

[0116] Multi-objective optimization (objectives: instantaneous sensory score, durability t) 1 / 2 (Skin compatibility score), using a weighted sum method:

[0117]

[0118] The weight w is obtained through group preference negotiation.

[0119] The output is an interpretable report, including the overall clustering performance score J for each group, the feature importance (βcoeff), and improvement suggestions (e.g., "increase the hydrophilic polymer by 0.05% to improve t"). 1 / 2 5%–10%”).

[0120] Results: Instantaneous subjective rating S_inst (mean, 0–10): Example group 7.2; Comparative group 6.6 (improvement +9.1%).

[0121] Durability half-life t 1 / 2 (At 32°C and after 10 cycles of friction): Example 1: 198 s; Comparative Example: 170 s (increase of +16.5%).

[0122] Skin compatibility score (0–100): Example 1 92; Comparative Example 89 (improvement +3.4%).

[0123] Overall clustering performance score J (weighted by the younger population): Example 1 0.78; Comparative Example 0.70 (relative improvement of 11.4%).

[0124] t 1 / 2 From the fitted λ, if λ = 0.0035 s -1 , then t 1 / 2 =ln2 / 0.0035≈198 s; J is calculated by substituting the normalized value into the above formula.

[0125] As shown above, this application, by applying acceleration perturbations and fitting response kinetics in controlled short-term experiments, can amplify effects that only manifest in long-term use scenarios within the experimental timescale. Coupled with these acceleration kinetics and non-invasive physiological indicators and characterized by an interpretable model, transferable predictions from short-term experiments to different real-world use scenarios can be achieved (i.e., mapping the kinetic laws obtained from accelerated experiments to expected performance under real-world conditions). Therefore, this evaluation is no longer merely an immediate description, but becomes a predictive tool based on the coupling of experimental kinetics and physiology, which helps in screening formulations with long-term robust performance and physiological compatibility in the early stages of research and development.

[0126] Example 3:

[0127] This embodiment is specifically applied to tinted sunscreen foundation (target: instant coverage + long-lasting makeup + oil control):

[0128] Sample B: Contains micro-dispersed titanium dioxide coloring particles, silicone oil-based oil control agent, and hydrophilic / hydrophobic balancing agent, aiming to improve instantaneous coverage and oil suppression duration.

[0129] Comparison (commercially available reference): Regular foundation, without special oil-controlling microparticles.

[0130] Step 2.1 (Obtaining the physical optical fingerprint):

[0131] Additional measurement of the coverage index COV (coverage) was determined using the color difference ΔE* on a standard background plate. For sample B, COV(ΔE* at t=0) was recorded as 12.5. Viscosity η(100 s⁻¹) -1 =900 mPa·s; particle size d50 = 0.28 μm.

[0132] Step 2.2 (obtaining short-term sensory responses):

[0133] The subjective short scale focuses on: immediate coverage (0–10), matte finish (0–10), and skin fit (0–10). Sampling points are the same as above, 0–600 s, video 120 fps, audio 16 kHz.

[0134] Step 2.3 (Obtaining temporal texture and optical variation description):

[0135] Masking decay ΔCOV(t) is extracted from the video (based on color difference over time), and oil film formation index OI(t) is extracted from the macro image (calculated using a high-contrast imaging algorithm). D includes the changes in ΔCOV, OI curves, and LBP.

[0136] Step 2.4 (Obtaining the accelerated response curve):

[0137] Controllable disturbance parameters: Simulated sweating (humidified air 40% relative humidity, local water mist spray 0.5 mL / cm²) 2 ) and friction (10 times). In this scenario, an exponential / polynomial fit is performed on ΔCOV(t):

[0138] ΔCOV(t)=C0 e−λcovert+C∞

[0139] We obtain λ_cover, and then the persistence half-life of the cover.

[0140] Step 2.5 (Establishing a sensory-physiological coupling model):

[0141] pH, skin temperature, a*, and sebum levels were collected (obtained by the weight difference method before and after blotting paper, unit: μg / cm³). 2 (as part of the non-invasive indicator G).

[0142] Step 2.6 (Clustered Scoring and Evaluation Output):

[0143] In the group preference, the oily group has a large weight for the matte feel. The score value of each group is obtained through multi-objective optimization and a report is generated.

[0144] Results (Example 3)

[0145] Instantaneous occlusion score S_inst (0–10):

[0146] Example 3, Group 8.0; Comparative Example 7.4 (improvement +8.1%).

[0147] Coverage half-life t 1 / 2,cover (Sweating + friction scenario):

[0148] Example 3: 240 s; Comparative Example: 205 s (improvement +17.1%).

[0149] Oil-controlling effect (percentage reduction in sebum after 60 minutes):

[0150] Example 3 showed an average reduction of 18%; the comparative example showed a reduction of 12% (an absolute increase of 6 percentage points and a relative increase of 50%).

[0151] Skin compatibility score (0–100):

[0152] Example 3, 90; Comparative Example 88 (increase +2.3%).

[0153] Overall clustering performance score J (using the oily group weight as an example): Example 3 0.82; Comparative Example 0.74 (relative improvement of 10.8%).

[0154] The comprehensive comparison is shown in Table 1 below (Final: Example 1, Example 2, Comparative Example - Key Indicators):

[0155] Indicators (units) Example 2 (Moisturizing Microemulsion) Example 3 (Tinted Foundation) Comparison ratio (commercially available reference) Instant subjective rating Sinst (0–10) 7.2 ± 0.6 8.0 ± 0.5 6.6 ± 0.7 <![CDATA[Durable half-life t 1 / 2 (seconds, 32 °C, 10 rubs)]]> 198 ± 22 240 ± 28 (coverage half-life) 170 ± 20 Skin compatibility score (0–100) 92 ± 4 90 ± 5 89 ± 5 Oil suppression effect (% reduction, 60 min) 12% ± 3% 18% ± 4% 12% ± 4% Cluster-based comprehensive score J (with group weight, 0–1) 0.78 0.82 0.72 Key Explainable Factors (Top 2) ΔI_mean (luminance decay), ΔpH OI (Oil Film Index), Δsebum No significant specific factors

[0156] As shown in Table 1 above, the method of this application couples physical optical fingerprints, multimodal temporal subjective and objective data, microstructure images, short-term acceleration perturbations and non-invasive physiological indicators to form an interpretable sensory-physiological coupling model. Based on group preferences, it performs multi-objective optimization and can simultaneously provide objective durability indicators and personalized scores and improvement suggestions for different user groups.

[0157] Experimental data from the two embodiments above show that the method achieves an 8–17% improvement in immediate sensory evaluation, durability, and overall grouping scores compared to the control group, while maintaining or slightly improving skin compatibility. This indicates that the method has practical value in actual evaluation and product iteration guidance.

[0158] As shown above, this application combines the output of the interpretable coupled model with hierarchical consumer preferences for multi-objective optimization, enabling the evaluation results to not only characterize general performance but also directly reflect the differences in preferences and weights of different user groups for performance dimensions. This segmented scoring system provides a clear decision-making path for R&D: for a specific target segment, the model can point out the most effective improvement direction and influencing factors, thereby shortening the closed-loop time from problem identification to formulation iteration. Compared to a single overall score, this mechanism makes the evaluation results more operable and economically reasonable in terms of commercial positioning, iteration priority setting, and small-batch trial production.

[0159] Example 4:

[0160] like Figure 2 As shown, another aspect of the present invention is to provide an evaluation device for cosmetics, comprising:

[0161] The physical optical fingerprint acquisition module is used to acquire cosmetic samples and subject baseline information, and to perform initial physical and optical characterization of the cosmetic samples under standardized light and temperature and humidity conditions to obtain physical optical fingerprints;

[0162] The short-term sensory and multimodal acquisition module is used to sample and simultaneously acquire high frame rate video, micro-audio, and subjective short scale scores to perform synchronous time synchronization, noise reduction, baseline removal, and standardization on the acquired raw time signals to obtain short-term sensory response time series.

[0163] The temporal texture and microstructure extraction module is used to perform spatial-temporal registration and alignment between the short-term sensory response time series and the macro skin images acquired in the same time window. After alignment, the spatiotemporal texture and optical evolution features are extracted based on operators of optical flow, local binary mode, and short-time Fourier transform to generate a temporal texture and optical change description.

[0164] The acceleration perturbation and response fitting module is used to superimpose the temporal texture and optical change description with the preset controllable perturbation parameters according to the experimental design to execute short-term acceleration scenarios, and to perform function fitting on the temporal response under each scenario to calculate and output the acceleration response curve and its dynamic parameters.

[0165] An interpretable coupling modeling module is used to input the accelerated response curve and the synchronously acquired non-invasive skin short-spectrum index into the causal discovery or variable screening process to screen candidate influence paths, and to fit the coupling relationship between sensory changes and microphysiological responses to establish an interpretable coupling model.

[0166] The cluster optimization and evaluation output module is used to map the output of the interpretable coupled model to a hierarchical consumer preference database, calculate the performance score of each cluster through clustering and multi-objective optimization, and summarize the performance score, feature importance and targeted improvement suggestions in the form of an interpretable report to output the clustered performance score and the final evaluation report.

[0167] In one embodiment, this application also provides an electronic device, the electronic device comprising:

[0168] At least one processor; and a memory communicatively connected to said at least one processor; wherein:

[0169] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described above.

[0170] In one embodiment, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0171] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for evaluating a cosmetic product, characterized by, The method comprises the following steps: obtaining cosmetic sample and baseline information of the subject, performing initial physical and optical characterization on the cosmetic sample under standardized lighting and temperature and humidity conditions to obtain a physical and optical fingerprint; binding the physical and optical fingerprint to a preset short scale, applying the sample to the local skin of the subject according to a uniform time sequence and simultaneously collecting high-frame-rate video, micro-audio and subjective short scale scores, synchronously correcting the time, denoising, removing the baseline and normalizing the collected time sequence of original signals to obtain a short-term sensory response time sequence; spatial-temporal registration and alignment of the short-term sensory response time sequence and the macro skin image collected at the same time window, extracting spatio-temporal texture and optical evolution features based on optical flow, local binary pattern and short-time Fourier transform operator after alignment to generate a time sequence texture and optical change description; superimposing the time sequence texture and optical change description and the preset controllable disturbance parameters according to the test design to perform short-time accelerated scenarios, and fitting the time sequence response under each scenario to calculate and output an accelerated response curve and its kinetic parameters; inputting the accelerated response curve and the synchronously collected non-invasive skin short spectrum index into a causal discovery or variable screening process to screen candidate influence paths, and fitting the coupling relationship between sensory changes and microphysiological reactions to establish an interpretable coupling model; mapping the output of the interpretable coupling model to a hierarchical consumer preference database, calculating the performance score of each cluster through clustering and multi-objective optimization, and summarizing the performance score, feature importance and targeted improvement suggestions in an interpretable report form to output the cluster performance score and the final evaluation report.

2. The cosmetic evaluation method according to claim 1, wherein the cosmetic sample and baseline information of the subject are obtained, and the cosmetic sample is initially physically and optically characterized under standardized lighting and temperature and humidity conditions to obtain a physical and optical fingerprint, comprising: using a spectrophotometer and a rotational viscometer to obtain a reflectance spectrum R(λ) and a viscosity curve η(·), and using dynamic light scattering to measure the particle size distribution to obtain a particle size d50, to perform initial physical and optical characterization to obtain a physical and optical fingerprint; the physical and optical fingerprint is obtained by principal component dimension reduction of the baseline corrected and normalized features.

3. The cosmetic evaluation method according to claim 1, wherein the physical and optical fingerprint is bound to a preset short scale, and the sample is applied to the local skin of the subject according to a uniform time sequence and simultaneously collects high-frame-rate video, micro-audio and subjective short scale scores, and the collected time sequence of original signals is synchronously corrected in time, denoised, baseline removed and normalized to obtain a short-term sensory response time sequence, comprising: ​ ​ The multi-modal acquisition at the subject's local skin site employs a high frame rate video acquisition frame rate no less than 120 fps and an audio sampling rate no less than 16 kHz, a subjective short-form scale is recorded at pre-set sampling points t={0, 30, 60, 120, 300, 600 s}, and the original time series data collected is aligned with a unified time stamp, band-pass filtered, de-noised, and zero-drifted to obtain a short-term sensory response time series.

4. The cosmetic evaluation method of claim 1, wherein, the short-term sensory response time series is spatially and temporally registered and aligned with a macro skin image collected in the same time window, and after alignment, spatio-temporal texture and optical evolution features are extracted based on an optical flow, a local binary pattern, and an operator of short-time Fourier transform to generate a time series texture and optical change description, including: first performing time-space registration on the video frames, then extracting a surface displacement field u(x, y, t) based on an optical flow method, and calculating a local binary pattern histogram and a local contrast to generate a time series texture and optical change description; the time series texture and optical change description is a matrix representation of a local binary pattern difference value, a brightness change curve, and a particle visibility index aggregated by time window.

5. The cosmetic evaluation method of claim 1, wherein, the time series texture and optical change description is superimposed with a pre-set controllable perturbation parameter according to a test design to perform a short-time accelerated scenario, and a function fitting is performed on the time series response under each scenario to calculate and output an accelerated response curve and its kinetic parameters, including: a combination of at least three temperature levels and three friction times, and in each perturbation scenario, a non-linear least squares method is used to fit an exponential decay or a polynomial to obtain an accelerated response curve and its kinetic parameters, and the application of the perturbation on the time axis is recorded by event time stamp to align with the time series data.

6. The cosmetic evaluation method of claim 1, wherein, the accelerated response curve and the synchronously collected non-invasive skin short spectrum index are input into a causal discovery or variable screening process to screen candidate influence paths, and a coupling relationship between sensory changes and microphysiological responses is fitted to establish an interpretable coupling model, including: a constraint-based causal discovery algorithm is first applied to the candidate variable set to screen out candidate causal edges, then a generalized linear regression or Bayesian regression with L1 sparse regularization is used to fit to obtain the interpretable coupling model, and a regularization parameter is selected through cross-validation to output the effect coefficient, significance test result, and confidence interval of each input feature in the interpretable coupling model.

7. The method for evaluating a cosmetic product according to claim 1, characterized in that, the output of the interpretable coupling model is mapped to a hierarchical consumer preference database, the performance score of each cluster is calculated through clustering and multi-objective optimization, and the performance score, feature importance, and targeted improvement suggestions are summarized in an interpretable report form to output a clustered performance score and a final evaluation report, including: the hierarchical consumer preference database is clustered based on the silhouette coefficient to determine the number of clusters and the k-means clustering is used to obtain the clusters. For each cluster, the instant sensory score, durability indicator, and skin compatibility are used as optimization objectives, and a weighted sum method or a Pareto front analysis is used for multi-objective optimization to calculate the cluster performance score, and an interpretable coupling model-based actionable improvement suggestion and corresponding uncertainty assessment are outputted in priority order.

8. An evaluation device for cosmetics, characterized in that, The method comprises the following steps: A physical and optical fingerprint acquisition module is used to obtain a cosmetic sample and baseline information of a subject, and to perform initial physical and optical characterization on the cosmetic sample under standardized light and temperature and humidity conditions to obtain a physical and optical fingerprint; A short-term sensory and multi-modal acquisition module is used to simultaneously acquire high-frame-rate video, micro-audio, and subjective short-form scale scores, and to perform synchronization time correction, denoising, baseline rejection, and standardization processing on the acquired time-series raw signals to obtain a short-term sensory response time series; A time-series texture and microstructure extraction module is used to spatially and temporally register and align the short-term sensory response time series with micro-distance skin images acquired at the same time window, and to extract spatio-temporal texture and optical evolution features based on optical flow, local binary pattern, and short-time Fourier transform operators to generate time-series texture and optical change descriptions; An accelerated disturbance and response fitting module is used to superimpose the time-series texture and optical change descriptions with preset controllable disturbance parameters according to a test design to perform short-time accelerated scenarios, and to perform function fitting on the time-series responses under each scenario to calculate and output an accelerated response curve and its kinetic parameters; An interpretable coupling modeling module is used to input the accelerated response curve and the simultaneously acquired non-invasive skin short-spectrum indicators into a causal discovery or variable screening process to screen candidate influence paths, and to fit the coupling relationship between sensory changes and microphysiological reactions to establish an interpretable coupling model; A cluster optimization and evaluation output module is used to map the output of the interpretable coupling model to a hierarchical consumer preference database, to calculate the performance score of each cluster through clustering and multi-objective optimization, and to summarize the performance score, feature importance, and targeted improvement suggestions in an interpretable report form to output a cluster performance score and a final evaluation report.

9. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein: The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The medium is a computer-readable storage medium, and the medium stores a computer program, and the computer program is executed by a processor to implement the method of any one of claims 1-7.

Images