Cosmetic raw material stability evaluation method based on multi-modal data fusion

Abstract

This invention relates to the field of cosmetic raw material testing technology, specifically to a method for evaluating the stability of cosmetic raw materials based on multimodal data fusion. The method includes the following steps: collecting multimodal data from a control group and an experimental group at multiple preset time points during the cultivation period to obtain a standardized dataset; internally fusing the same modality data within the standardized dataset to extract the core features of each modality, and performing cross-modal fusion of the core features of different modalities within the standardized dataset to obtain a fused feature vector; obtaining the sample stability level and key influencing factors from the fused feature vector, comparing the evaluation results of the control group and the experimental group, and outputting a final stability evaluation report. Through this method, multimodal data can be integrated to achieve accurate, comprehensive, and efficient evaluation, thereby filling a technological gap in the industry and helping the cosmetic industry achieve scientific and digital control of raw material quality.

Description

Technical Field

[0001] This invention relates to the field of cosmetic raw material testing technology, and in particular to a method for evaluating the stability of cosmetic raw materials based on multimodal data fusion. Background Technology

[0002] With the rapid development of the cosmetics industry, consumers are paying unprecedented attention to product quality and safety. The stability of cosmetic raw materials, as the core intrinsic attribute that determines the final product quality, efficacy, and safety of use, is directly related to the product's shelf life, consumer experience, and market competitiveness, and plays a fundamental role in the technological development of the cosmetics industry.

[0003] Currently, the methods for assessing the stability of raw materials in the cosmetics industry are relatively traditional, mainly employing single-modal detection or single-dimensional evaluation. Existing publicly available assessment methods also have significant limitations, failing to meet the industry's demands for high-quality development. For example, some assessment methods only extract color parameters from collected images to achieve digital evaluation of stability, reflecting only changes in the raw material's appearance color and failing to comprehensively characterize the overall stability. Other methods use single physicochemical index detection (such as acid value or viscosity) or sensory evaluation. Physicochemical detection only reflects changes in a specific physicochemical property, while sensory evaluation relies on subjective human judgment, easily influenced by individual differences and visual fatigue, resulting in large subjective errors and poor repeatability. Still other detection methods (such as particle size analysis and light scattering) suffer from drawbacks such as being highly destructive, insensitive to minor instability phenomena, or having high equipment costs and limited applicability, making them unsuitable for rapid and accurate assessment of large batches of raw materials.

[0004] In summary, existing single-modal evaluation methods cannot adapt to the evaluation needs of new raw materials, cannot meet the actual needs of enterprises for precise control of raw material quality, and cannot align with the development trend of the cosmetics industry towards digitalization and intelligentization. Summary of the Invention

[0005] The purpose of this invention is to provide a method for evaluating the stability of cosmetic raw materials based on multimodal data fusion. This method aims to solve the technical problems that existing single-modal evaluation methods cannot adapt to the evaluation needs of new raw materials, cannot meet the actual needs of enterprises for precise control of raw material quality, and cannot meet the development trend of the cosmetic industry towards digitalization and intelligence.

[0006] To achieve the above objectives, this invention employs a method for evaluating the stability of cosmetic raw materials based on multimodal data fusion, comprising the following steps: The cosmetic raw materials to be evaluated were purified and packaged into parallel samples, and divided into control group and experimental group for cultivation. At multiple preset time points during the cultivation period, multimodal data of physicochemical, image, sensory and environmental response of control group and experimental group were collected. The collected multimodal data were standardized, outlier removed and missing value filled to obtain standardized dataset. A hierarchical fusion approach is adopted to perform internal fusion of data of the same modality in the standardized dataset, extract the core features of each modality, and perform cross-modal fusion of the core features of different modalities in the standardized dataset to obtain a fused feature vector. The stability level and key influencing factors of the sample are obtained by fusion feature vectors, and the evaluation results of the control group and the experimental group are compared to output the final stability evaluation report.

[0007] The process includes the following steps: First, the raw materials for the cosmetics to be evaluated are purified, packaged into parallel samples, and divided into control and experimental groups for cultivation. During the cultivation period, multimodal data on physicochemical properties, images, sensory responses, and environmental responses are collected from both control and experimental groups at multiple pre-set time points. Then, the collected multimodal data is standardized, outlier removal is performed, and missing values ​​are filled in to obtain a standardized dataset. Select cosmetic raw materials to be evaluated, remove impurities from the raw materials, and then divide them into multiple parallel samples. All parallel samples are divided into control group and experimental group. The control group samples were cultured under standard environmental conditions, while the experimental group samples were cultured under different stress environmental conditions, with a uniform preset culture time. Within the preset culture period, multiple fixed time points were selected to collect multimodal data from all samples in the control group and the experimental group. The multimodal data included four categories: physicochemical characteristic data, image characteristic data, sensory characteristic data, and environmental response data.

[0008] Within a pre-defined culture period, multiple fixed time points were selected to collect multimodal data from all samples in both the control and experimental groups. This multimodal data included four categories: physicochemical characteristic data, image characteristic data, sensory characteristic data, and environmental response data. Preprocess the multimodal data.

[0009] Among the steps in preprocessing multimodal data: A unified standardization method is used to eliminate the dimensional differences between different types of data, and outlier data is removed by outlier identification method. Missing items in the data are filled by missing value completion method, and finally a standardized dataset is obtained.

[0010] In the steps of using a hierarchical fusion approach to internally fuse data of the same modality in a standardized dataset, extracting the core features of each modality, and then performing cross-modal fusion of the core features of different modalities in the standardized dataset to obtain a fused feature vector: For the four types of multimodal data in the standardized dataset, internal fusion processing is performed on each type of single modality data. Dimensionality reduction method is used to simplify the single modality data and select key data that reflects the core information of the modality. Core features are extracted from the fused data within each type of single-modality data. A cross-modal fusion method is adopted to integrate the core features of all single modalities.

[0011] In the step of extracting core features from the fused data within each type of single-modality data: By retaining the characteristic information reflecting the stability of cosmetic raw materials and eliminating irrelevant and redundant features, the core features of each individual modality are obtained.

[0012] After the step of integrating the core features of all single modalities using a cross-modal fusion method: Through fusion processing, the core features of different modalities are integrated into a unified feature vector, i.e., the fused feature vector, and then output.

[0013] Among them, the steps of obtaining the sample stability level and key influencing factors based on the fused feature vectors, comparing the evaluation results of the control group and the experimental group, and outputting the final stability evaluation report are as follows: The stability level and key factors affecting the stability of each sample are obtained based on the fused feature vector. The stability assessment results of the control group samples and the experimental group samples were compared one by one to analyze the impact of different culture environments on the stability of cosmetic raw materials.

[0014] After the steps of obtaining the stability level of each sample and the key factors affecting the stability of that sample based on the fused feature vector: The stability levels were divided into four categories: excellent, good, acceptable, and unacceptable. At the same time, the specific types and degrees of influence of key influencing factors were also identified.

[0015] The process includes comparing the stability assessment results of the control group and the experimental group samples one by one to analyze the impact of different culture environments on the stability of cosmetic raw materials: The evaluation data, comparison results, stability levels, and key influencing factors of all samples are integrated to form a complete stability evaluation report.

[0016] This invention discloses a method for assessing the stability of cosmetic raw materials based on multimodal data fusion. The method involves removing impurities from the cosmetic raw materials to be assessed, packaging them into parallel samples, and dividing them into control and experimental groups for cultivation. During cultivation, physicochemical, image, sensory, and environmental response multimodal data are collected from both control and experimental groups at multiple preset time points. The collected multimodal data is then standardized, outlier removal is performed, and missing values ​​are filled to obtain a standardized dataset. A hierarchical fusion approach is used to internally fuse data of the same modality within the standardized dataset, extracting the core features of each modality. Furthermore, cross-modal fusion is performed on the core features of different modalities within the standardized dataset to obtain a fused feature vector. The stability level and key influencing factors of the samples are obtained from the fused feature vector, and the assessment results of the control and experimental groups are compared to output a final stability assessment report. This method integrates multimodal data to achieve accurate, comprehensive, and efficient assessment, thereby filling a technological gap in the industry and helping the cosmetic industry achieve scientific and digital control of raw material quality. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the steps of the cosmetic raw material stability assessment method based on multimodal data fusion of the present invention.

[0019] Figure 2 This is a flowchart of steps S100 of the present invention.

[0020] Figure 3 This is a flowchart of steps S200 of the present invention.

[0021] Figure 4 This is a flowchart of steps S300 of the present invention.

[0022] Figure 5 This is a schematic diagram of the structural principle of the cosmetic raw material stability assessment system based on multimodal data fusion of the present invention.

[0023] Figure 6 This is a schematic diagram of the electronic device of the present invention.

[0024] 401 - Multimodal data acquisition module, 402 - Multimodal data hierarchical fusion module, 403 - Stability assessment module. Detailed Implementation

[0025] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0026] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0027] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0028] Please see Figures 1-4 This invention provides a method for evaluating the stability of cosmetic raw materials based on multimodal data fusion, comprising the following steps: S100: The cosmetic raw materials to be evaluated are purified and packaged into parallel samples, divided into control and experimental groups for incubation. At multiple preset time points during the incubation period, multimodal data on physicochemical properties, images, sensory responses, and environmental responses are collected from both control and experimental groups. The collected multimodal data are then standardized, outlier removed, and missing value imputed to obtain a standardized dataset. The specific process is as follows: S101: Select the cosmetic raw materials to be evaluated, remove impurities from the raw materials, and then divide them into multiple parallel samples. All parallel samples are divided into control group and experimental group. S102: The control group samples were cultured under standard environmental conditions, and the experimental group samples were cultured under different stress environmental conditions, with a uniform preset culture time set. S103: Within the preset culture period, select multiple fixed time points to collect multimodal data from all samples in the control group and the experimental group respectively; the multimodal data includes four categories: physicochemical characteristic data, image characteristic data, sensory characteristic data, and environmental response data; S104: Preprocess the multimodal data, use a unified standardization method to eliminate the dimensional differences between different types of data, remove outlier data through outlier identification methods, and fill in missing items in the data using missing value completion methods to finally obtain a standardized dataset.

[0029] In the above process, the cosmetic raw materials to be evaluated are selected, impurities are removed, and the raw materials are divided into multiple parallel samples. All parallel samples are then divided into a control group and an experimental group. Specifically, the cosmetic raw materials to be evaluated that are free from deterioration, damage, and uniform composition are first screened. Solid impurities, foreign objects, and unqualified particles in the raw materials are removed by filtration, centrifugation, or manual sorting to ensure that the purity of the raw materials meets the evaluation requirements. Subsequently, according to a uniform preset specification (e.g., 40~50mL / g per sample), the processed raw materials are divided into multiple parallel samples, with no fewer than 3 parallel samples to ensure data repeatability. Finally, a random grouping method is used to divide all parallel samples equally into a control group and an experimental group, ensuring that the specifications, quantity, and initial state of the two groups of samples are completely consistent to avoid interference from group differences in subsequent evaluation results.

[0030] The control group samples were cultured under standard environmental conditions, while the experimental group samples were cultured under different stress environmental conditions, with a uniform preset culture time. Specifically, the control group was cultured in a standard environment, i.e., temperature 25±2℃, relative humidity 50±5%, and no light, with environmental parameters maintained constant using a stability test chamber. The experimental groups were placed in different stress environments, including high temperature environments (40±2℃, 50±2℃), low temperature environments (-10±2℃), and light environments (UV irradiation, intensity 30±5W / m²). 2 High humidity environment (relative humidity 85±5%), each stress environment corresponds to a set of test samples, and the stability test chamber is used to accurately control each environmental parameter to avoid environmental fluctuations; the preset culture time is uniformly set to 30~90 days, and the environmental parameters are monitored in real time during the culture period to ensure that the culture conditions of each group are stable.

[0031] Within a pre-set culture period, multiple fixed time points were selected to collect multimodal data from all samples in both the control and experimental groups. The multimodal data included four categories: physicochemical characteristic data, image characteristic data, sensory characteristic data, and environmental response data. Specifically, the pre-set time points were rationally set according to the culture duration; for example, in a 60-day culture period, days 0, 15, 30, 45, and 60 could be selected. Data collection was conducted at the same time period for each point to avoid errors caused by time differences. The specific data collection process for each type of data is as follows: Physicochemical characteristic data were collected using corresponding testing instruments, including acid value, pH, and other parameters. Oxidation value, active ingredient content, pH value, viscosity, etc., with active ingredient content detected by high-performance liquid chromatography and heavy metal content detected by atomic absorption spectrophotometry; image feature data were collected using a high-definition camera (resolution not less than 1080P) under uniform white light background, shooting angle and distance to ensure consistent shooting conditions; sensory feature data were evaluated by an evaluation team of 3 or more professional evaluators according to a unified scoring standard; environmental response data were collected in real time through environmental sensors, synchronously recording the temperature, humidity, light intensity of the sample culture environment and the sample's own environmental response parameters.

[0032] Multimodal data undergoes preprocessing. A unified standardization method is used to eliminate dimensional differences between different data types. Outlier identification is employed to remove outliers, and missing value imputation is used to fill in missing data, ultimately resulting in a standardized dataset. The specific processing steps are as follows: Standardization uses the Z-score standardization method to convert various types of multimodal data into standardized data with a mean of 0 and a standard deviation of 1, eliminating dimensional differences between different data types (such as viscosity, pH value, and rating) and ensuring data comparability. Outlier identification uses the 3σ principle, calculating the mean and standard deviation of each type of data. Data exceeding the mean ± 3 times the standard deviation are identified as outliers and removed to avoid interference with evaluation results. Missing value imputation uses the K-nearest neighbor algorithm, selecting multiple samples with the most similar features to the missing data samples and filling in the missing items by calculating their mean, ensuring the integrity of the dataset. All preprocessing steps are performed according to a unified standard. After preprocessing, the dataset is integrated to obtain a standardized and complete dataset for subsequent fusion processing.

[0033] S200: Employs a hierarchical fusion approach, internally fusing data of the same modality within a standardized dataset to extract core features from each modality, and then performing cross-modal fusion of core features from different modalities within the standardized dataset to obtain a fused feature vector. The specific process is as follows: S201: For the four types of multimodal data in the standardized dataset, internal fusion processing is performed on each type of single modality data. Dimensionality reduction method is used to simplify the single modality data and select key data that reflects the core information of the modality. S202: Extract core features from the data after fusion within each single modality, retain the feature information reflecting the stability of cosmetic raw materials, and remove irrelevant and redundant features to obtain the core features of each single modality. S203: Employs a cross-modal fusion method to integrate the core features of all single modalities; S204: Through fusion processing, the core features of different modalities are integrated into a unified feature vector, i.e., the fused feature vector, and then output.

[0034] In the above process, for the four types of multimodal data in the standardized dataset, internal fusion processing was performed on each type of single-modal data. Dimensionality reduction methods were used to simplify the single-modal data and select key data reflecting the core information of the modality. Specifically, internal fusion was performed on the four types of single-modal data: physicochemical characteristics, image features, sensory characteristics, and environmental response data. Principal component analysis was used as a dimensionality reduction method in the fusion process. Dimensionality reduction processing was performed on the standardized data of each modality. By calculating eigenvalues ​​and eigenvectors, key data with larger eigenvalues ​​that could reflect the core information of the modality were selected, redundant data were eliminated, and the simplification of single-modal data was achieved, while retaining the core information of the modality related to the stability of cosmetic raw materials.

[0035] Core features are extracted from the fused data within each single modality, retaining features reflecting the stability of cosmetic raw materials while removing irrelevant and redundant features to obtain the core features of each single modality. Specifically, for the simplified data fused within each single modality, a feature selection algorithm is used, combined with factors influencing the stability of cosmetic raw materials, to select features highly correlated with stability, such as acid value and peroxide value in physicochemical features, color change parameters in image features, texture scores in sensory features, and temperature response rate in environmental response data. Redundant features unrelated to stability are removed to ensure that the extracted core features accurately reflect the changes in raw material stability under that modality, ultimately yielding the core features for each of the four single modalities. Among these: The feature selection algorithm used is the Relief-F algorithm. This algorithm is suitable for feature selection requirements of multimodal data and can accurately select core features that are highly related to the stability of cosmetic raw materials, while eliminating irrelevant and redundant features. The specific implementation process is as follows: Initialization parameters: Set the number of nearest neighbor samples k (the value is 5~10, determined in combination with the number of cosmetic raw material samples, and k=8 is uniformly set in this evaluation method), the feature weight threshold θ (the value is 0.05, used to determine whether the feature is retained), and the number of iterations is consistent with the number of samples in the standardized dataset to ensure that each sample can participate in the feature weight calculation.

[0036] Sample classification and nearest neighbor selection: The single-modal data (physicochemical feature data, image feature data, sensory feature data, and environmental response data) in the standardized dataset are screened separately and classified according to the preliminary stability judgment results of the raw materials (stable, relatively stable, and unstable categories based on the initial test data), resulting in a set of similar samples and a set of dissimilar samples. For each sample, k nearest neighbors are selected from the set of similar samples as similar neighbors, and k nearest neighbors are selected from the set of dissimilar samples as dissimilar neighbors. Euclidean distance is used to calculate the distance to ensure the representativeness of the nearest neighbor samples.

[0037] Feature weight update: Iterate through all features of each sample, calculate the average difference between the feature and the current sample in the same category, and the average difference between the feature and the current sample in the opposite category; update the weight of the feature based on the difference value. If the difference between the feature and the current sample in the opposite category is larger and the difference between the feature and the current sample in the same category is smaller, it means that the feature has a higher degree of distinguishing the stability of raw materials, and the weight value is larger; conversely, the weight value is smaller, until the weight update of all samples and features is completed.

[0038] Feature selection and output: After iteration, the final weight values ​​of all features are obtained. Features with weight values ​​greater than the set threshold θ are retained as the core features of the modality. Features with weight values ​​less than or equal to the threshold θ are judged as irrelevant or redundant features and are removed. Finally, the core feature set of each single modality (physicochemical, image, sensory, environmental response) is output to ensure that the core features can accurately reflect the change law of the stability of cosmetic raw materials and provide high-quality feature support for subsequent cross-modal fusion.

[0039] For different modal data, the parameters of the feature selection algorithm can be fine-tuned. For example, the weight threshold for image feature data can be adjusted to 0.04 to ensure that key features such as color changes and texture changes are not missed; the weight threshold for physicochemical feature data is maintained at 0.05 to retain core physicochemical indicators such as acid value, peroxide value, and active ingredient content.

[0040] A cross-modal fusion method is adopted to integrate the core features of all single modalities. Specifically, an attention mechanism is used as the cross-modal fusion method to integrate the core features of four modalities: physicochemical features, image features, sensory features, and environmental response data. Through an adaptive learning algorithm, different attention weights are assigned according to the degree of influence of each modal core feature on the stability of cosmetic raw materials. Modal features with a greater impact on stability are assigned higher weights, while modal features with a smaller impact are assigned lower weights, highlighting the role of key features while taking into account the integrity of features of various modalities, thus achieving effective integration of core features of different modalities. The specific implementation process is as follows: Feature preprocessing: The four types of modal core features (physicochemical feature core set, image feature core set, sensory feature core set, and environmental response feature core set) extracted in step S202 are subjected to dimensional unification processing. A linear mapping method is used to map the core features of each modality to the same dimensional space (the dimension is set to 64 dimensions, which can be fine-tuned according to the number of samples and feature complexity), eliminating the dimensional differences of core features of different modalities and ensuring that fusion calculation can be performed.

[0041] Attention weight calculation: Construct a channel attention module to assign weights to the core features of the four modalities after unifying the dimensions. This is specifically divided into three sub-steps: (1) Global average pooling: Global average pooling is performed on the core features of each modality to compress the multi-dimensional core features of each modality into a 1-dimensional feature vector. This vector can represent the overall information of the core features of the modality and reflect the overall influence of the modality on the stability of the raw materials.

[0042] (2) Fully connected mapping: The four 1-dimensional feature vectors (corresponding to the four modalities) obtained by global average pooling are input into the fully connected layer and two linear mappings are performed. The first mapping compresses the feature dimension to 16 dimensions, and the second mapping restores it to 4 dimensions. The activation function (using the Sigmoid function) outputs four weight values ​​between 0 and 1, which correspond to the attention weights of the core features of the four modalities.

[0043] (3) Weight calibration: Based on prior knowledge of the stability assessment of cosmetic raw materials, the weight values ​​output by the fully connected layer are calibrated. For example, the physicochemical characteristics and environmental response characteristics usually have a greater impact on the stability of raw materials. Their initial weight values ​​can be multiplied by a calibration coefficient of 1.2, and the sensory characteristics weight values ​​can be multiplied by a calibration coefficient of 0.9 to ensure that the weight allocation meets the actual assessment requirements and avoid weight allocation deviation.

[0044] Cross-modal fusion computation: The four modal core features after unifying the dimensions are multiplied by the corresponding attention weights to obtain the weighted modal core features; then, all the weighted modal core features are added element by element to obtain a 64-dimensional unified feature vector, i.e., the fusion feature vector.

[0045] Feature Optimization: The obtained fusion feature vector is normalized by using the Min-Max normalization method to map each element in the vector to between 0 and 1, eliminating the influence of dimensions; at the same time, redundant elements with values ​​close to 0 in the fusion feature vector are removed to further optimize the effectiveness of the fusion features. Finally, the optimized fusion feature vector is output for the stability evaluation of the subsequent step S300.

[0046] The weight allocation of the attention mechanism has adaptive learning capabilities. It can be trained with a large amount of cosmetic raw material sample data to continuously optimize the weight calibration coefficient, making the weight allocation more in line with the stability assessment needs of different types of cosmetic raw materials and improving the accuracy of the fused feature vector.

[0047] Through fusion processing, the core features of different modalities are integrated into a unified feature vector, namely the fused feature vector, and output. The specific processing procedure is as follows: based on the attention weights allocated in step S203, the core features of various modalities are weighted and fused to integrate the multi-dimensional core features into a feature vector with unified dimensions and comprehensive information, namely the fused feature vector. After fusion, the fused feature vector is normalized to ensure that the vector data is standardized, and then the fused feature vector is output for subsequent stability evaluation steps.

[0048] S300: This function obtains the sample stability level and key influencing factors from the fused feature vectors, compares the evaluation results of the control group and the experimental group, and outputs a final stability evaluation report. The specific process is as follows: S301: Obtain the stability level and key factors affecting the stability of each sample based on the fused feature vector; S302: Classify the stability levels, dividing the sample stability into four levels: excellent, good, qualified, and unqualified, and also classify the specific types and degree of influence of key influencing factors; S303: Compare the stability assessment results of the control group samples and the experimental group samples one by one to analyze the impact of different culture environments on the stability of cosmetic raw materials; S304: Integrate all sample evaluation data, comparison results, stability levels, and key influencing factors to form a complete stability assessment report.

[0049] In the above process, the stability level and key factors affecting the stability of each sample are obtained based on the fusion feature vector. Specifically, a deep learning-based fusion model (composed of a convolutional neural network (CNN) and a recurrent neural network (RNN)) is used to process the fusion feature vector. The generation process of this model is as follows: First, multimodal data of different types of cosmetic raw materials and corresponding actual stability test results are collected to construct training and test datasets. The training dataset is input into the initial fusion model, and the model parameters are continuously adjusted through the backpropagation algorithm to minimize the error between the model evaluation results and the actual detection results. The trained model is validated using the test dataset. When the model accuracy is ≥95%, it is determined as the final stability evaluation model. The fusion feature vector is input into the trained evaluation model. The model analyzes the multi-dimensional information in the fusion feature vector and outputs the stability level and key influencing factors corresponding to each sample. The key influencing factors include environmental factors (temperature, humidity, light) and raw material factors (active ingredient content, physicochemical properties).

[0050] The stability of samples was categorized into four levels: excellent, good, acceptable, and unacceptable. The specific types and degrees of influence of key influencing factors were also identified. Specifically, stability level classification standards were pre-defined, and thresholds for each level were set based on relevant standards in the cosmetic raw material industry and practical testing experience: Excellent level: no significant changes in raw materials, with all characteristic parameters stable within a reasonable range; Good level: slight changes in raw materials, not affecting safety and efficacy; Acceptable level: some changes in raw materials, but within safe limits; Unacceptable level: significant deterioration or degradation of raw materials, posing a safety hazard. Furthermore, key influencing factors were classified, clarifying the specific types of each factor (such as high temperature, light exposure, and active ingredient degradation), and the degree of influence of each factor on sample stability was analyzed using a model, classifying the influence level as strong, moderate, or weak.

[0051] The stability assessment results of the control group samples and the experimental group samples were compared one by one to analyze the impact of different culture environments on the stability of cosmetic raw materials. Specifically, a one-to-one comparison method was used to compare the stability level and key influencing factors of each sample in the control group with the assessment results of the corresponding samples in the experimental group under each stress environment. The focus was on analyzing the impact of different stress environments (high temperature, low temperature, light, etc.) on the stability level of raw materials compared with the standard environment of the control group, as well as the differences in key influencing factors under different environments, to clarify the influence patterns and extent of various environmental factors on the stability of cosmetic raw materials.

[0052] A comprehensive stability assessment report is generated by integrating all sample evaluation data, comparison results, stability levels, and key influencing factors. Specifically, the process involves first reviewing the basic information of all samples (raw material type, specifications, culture conditions, etc.), preprocessing results of multimodal data, and fusing feature vector information; then integrating the stability levels, key influencing factors, and classification criteria for each sample; subsequently summarizing the comparison results between the control and experimental groups to clarify the impact of different environments on raw material stability; and finally, compiling a complete stability assessment report. This report is clearly structured, data-rich, and comprehensively presents the assessment process and results, providing a reliable basis for the quality control and storage optimization of cosmetic raw materials.

[0053] Corresponding to the aforementioned embodiments of the cosmetic raw material stability assessment method based on multimodal data fusion, this application also provides embodiments of the cosmetic raw material stability assessment system based on multimodal data fusion.

[0054] Figure 5 This is a schematic diagram illustrating the structural principle of a cosmetic raw material stability assessment system based on multimodal data fusion, according to an exemplary embodiment. (Refer to...) Figure 5 The system may include: a multimodal data acquisition module 401, a multimodal data hierarchical fusion module 402, and a stability evaluation module 403; wherein: The multimodal data acquisition module 401 is used to remove impurities from the cosmetic raw materials to be evaluated, package them into parallel samples, divide them into control group and experimental group for cultivation, collect physicochemical, image, sensory and environmental response multimodal data of control group and experimental group at multiple preset time nodes during cultivation, and standardize, remove outliers and fill missing values ​​in the collected multimodal data to obtain a standardized dataset. The multimodal data hierarchical fusion module 402 is used to perform internal fusion of the same modality data in the standardized dataset using a hierarchical fusion method, extract the core features of each modality, and perform cross-modal fusion of the core features of different modalities in the standardized dataset to obtain a fused feature vector. The stability assessment module 403 is used to obtain the sample stability level and key influencing factors based on the fused feature vector, compare the assessment results of the control group and the experimental group, and output the final stability assessment report.

[0055] In this embodiment, the multimodal data acquisition module 401 removes impurities from the cosmetic raw materials to be evaluated and packages them into parallel samples, dividing them into control and experimental groups for cultivation. During the cultivation period, it collects physicochemical, image, sensory, and environmental response multimodal data from the control and experimental groups at multiple preset time points. The collected multimodal data is then standardized, outlier removal is performed, and missing values ​​are filled to obtain a standardized dataset. The multimodal data hierarchical fusion module 402 uses a hierarchical fusion method to internally fuse data of the same modality in the standardized dataset, extract the core features of each modality, and perform cross-modal fusion of the core features of different modalities in the standardized dataset to obtain a fused feature vector. The stability assessment module 403 obtains the sample stability level and key influencing factors based on the fused feature vector, compares the assessment results of the control and experimental groups, and outputs a final stability assessment report. Through the above methods, multimodal data can be integrated to achieve accurate, comprehensive, and efficient assessment, thereby filling the industry's technological gap and helping the cosmetic industry achieve scientific and digital control of raw material quality.

[0056] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0057] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0058] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the above-described method for evaluating the stability of cosmetic raw materials based on multimodal data fusion. Figure 6 The diagram shown is a hardware structure diagram of any device with data processing capabilities, which is part of a cosmetic raw material stability assessment system based on multimodal data fusion provided in an embodiment of the present invention. (Except for...) Figure 6 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0059] Accordingly, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the aforementioned method for evaluating the stability of cosmetic raw materials based on multimodal data fusion. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0060] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0061] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for evaluating the stability of cosmetic raw materials based on multimodal data fusion, characterized in that, Includes the following steps: The cosmetic raw materials to be evaluated were purified and packaged into parallel samples, and divided into control group and experimental group for cultivation. At multiple preset time points during the cultivation period, multimodal data of physicochemical, image, sensory and environmental response of control group and experimental group were collected. The collected multimodal data were standardized, outlier removed and missing value filled to obtain standardized dataset. A hierarchical fusion approach is adopted to perform internal fusion of data of the same modality in the standardized dataset, extract the core features of each modality, and perform cross-modal fusion of the core features of different modalities in the standardized dataset to obtain a fused feature vector. The stability level and key influencing factors of the sample are obtained by fusion feature vectors, and the evaluation results of the control group and the experimental group are compared to output the final stability evaluation report.

2. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 1, characterized in that, The process involves removing impurities from the cosmetic raw materials to be evaluated, packaging them into parallel samples, dividing them into control and experimental groups for cultivation, collecting multimodal data on physicochemical properties, images, sensory responses, and environmental responses from both groups at multiple pre-set time points during cultivation, and standardizing, removing outliers, and imputing missing values ​​from the collected multimodal data to obtain a standardized dataset. Select cosmetic raw materials to be evaluated, remove impurities from the raw materials, and then divide them into multiple parallel samples. All parallel samples are divided into control group and experimental group. The control group samples were cultured under standard environmental conditions, while the experimental group samples were cultured under different stress environmental conditions, with a uniform preset culture time. Within the preset culture period, multiple fixed time points were selected to collect multimodal data from all samples in the control group and the experimental group. The multimodal data included four categories: physicochemical characteristic data, image characteristic data, sensory characteristic data, and environmental response data.

3. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 2, characterized in that, Within a pre-set culture period, multiple fixed time points were selected to collect multimodal data from all samples in both the control and experimental groups. This multimodal data included four categories: physicochemical characteristic data, image characteristic data, sensory characteristic data, and environmental response data. Preprocess the multimodal data.

4. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 3, characterized in that, In the preprocessing steps of multimodal data: A unified standardization method is used to eliminate the dimensional differences between different types of data, and outlier data is removed by outlier identification method. Missing items in the data are filled by missing value completion method, and finally a standardized dataset is obtained.

5. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 1, characterized in that, In the steps of using a hierarchical fusion approach to internally fuse data of the same modality in a standardized dataset, extract the core features of each modality, and perform cross-modal fusion of the core features of different modalities in the standardized dataset to obtain a fused feature vector: For the four types of multimodal data in the standardized dataset, internal fusion processing is performed on each type of single modality data. Dimensionality reduction method is used to simplify the single modality data and select key data that reflects the core information of the modality. Core features are extracted from the fused data within each type of single-modality data. A cross-modal fusion method is adopted to integrate the core features of all single modalities.

6. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 5, characterized in that, In the step of extracting core features from the fused data within each type of single-modality data: By retaining the characteristic information reflecting the stability of cosmetic raw materials and eliminating irrelevant and redundant features, the core features of each individual modality are obtained.

7. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 6, characterized in that, After employing a cross-modal fusion method to integrate the core features of all individual modalities: Through fusion processing, the core features of different modalities are integrated into a unified feature vector, i.e., the fused feature vector, and then output.

8. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 1, characterized in that, In the steps of obtaining the sample stability level and key influencing factors based on the fused feature vectors, comparing the evaluation results of the control group and the experimental group, and outputting the final stability evaluation report: The stability level and key factors affecting the stability of each sample are obtained based on the fused feature vector. The stability assessment results of the control group samples and the experimental group samples were compared one by one to analyze the impact of different culture environments on the stability of cosmetic raw materials.

9. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 8, characterized in that, After obtaining the stability level of each sample and the key factors affecting the stability of that sample based on the fused feature vector: The stability levels were divided into four categories: excellent, good, acceptable, and unacceptable. At the same time, the specific types and degrees of influence of key influencing factors were also identified.

10. The method for evaluating the stability of cosmetic raw materials based on multimodal data fusion as described in claim 9, characterized in that, After comparing the stability assessment results of the control group samples and the experimental group samples one by one to analyze the impact of different culture environments on the stability of cosmetic raw materials: The evaluation data, comparison results, stability levels, and key influencing factors of all samples are integrated to form a complete stability evaluation report.

Images