Multi-modal data fused marinated product flavor directional design method and system

Abstract

The invention belongs to the technical field of intelligent food manufacturing, and particularly discloses a multi-modal data fused marinated product flavor directional design method and system. S1, constructing an auxiliary material database; s2, mining internal associations of components in the high-quality auxiliary materials by adopting an Apriori algorithm; s3, evaluating the sensory flavor of the marinated product to obtain a product index data set; s4, establishing a multivariate mapping model between the auxiliary material parameters and the sensory quality on the basis of a Python library of Statsmodels; s5, establishing a user score matrix in combination with consumption preference; and S6, optimizing the auxiliary material component model by using a genetic algorithm, and solving by combining the multivariate mapping model obtained in the S4 and the sensory weight obtained in the S5 to obtain the auxiliary material component with the optimal quality. By adopting the multi-modal data fused marinated product flavor directional design method and system disclosed by the invention, the problems of sensory flavor quality consistency control limitation and auxiliary material optimization intelligent management design management of traditional sauce marinated poultry meat are solved, and intelligent control between auxiliary material and flavor sensory is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of food intelligent manufacturing, and in particular to a marinated product flavor directional design method and system fusing multi-modal data. BACKGROUND

[0002] Marinated products are processed by pre-cooking, soaking, cooking, marinating and other processes with fresh (frozen) poultry meat and edible by-products in water added with various auxiliary materials. The basic flavor of traditional sauce-marinated meat products is mainly realized by marinating process, and high-quality marinating auxiliary materials are the key to the flavor presentation of sauce-marinated poultry meat products. However, due to the lack of necessary quantitative standards for the quality of recycled marinating auxiliary materials, the quality of different batches of products such as taste and color is unstable.

[0003] With the continuous growth of market demand and the increasing improvement of consumer taste, the optimization management and directional flavor design of marinated product auxiliary materials have become a key problem to be solved in the industry. The formulation of traditional sauce-marinating auxiliary materials mainly relies on the process experience and sensory evaluation of marinating personnel. Through years of practical accumulation, the meat products are evaluated by means of olfaction, taste and other senses, and the best auxiliary material composition is determined by repeatedly adjusting the process according to the evaluation results and target flavor characteristics. Although the traditional marinating method ensures the quality of sauce-marinated poultry meat to a certain extent, it still has certain limitations. First, sensory evaluation is subjective and uncertain, and it is difficult to realize standardization and data traceability. Second, the formulation process of marinating auxiliary materials lacks quantitative indicators and data support, and it is difficult to achieve precise control and optimization. In recent years, with the development of emerging technologies such as big data and artificial intelligence, the use of multi-modal data-driven methods to optimize and manage sauce-marinated poultry meat auxiliary materials has received widespread attention in the industry.

[0004] In the prior art, researchers use machine learning algorithms such as neural networks to build an intelligent recommendation system for sauce-marinated meat products, and according to the target flavor characteristics and on-sale product data, the optimal process flow is predicted. These data-driven methods to a certain extent overcome the limitations of traditional sauce-marinated meat processing methods, and provide new ideas and approaches for sauce-marinated poultry meat auxiliary material optimization and flavor directional design technology.

[0005] However, the existing modal data-driven methods still have some deficiencies in practical application. On the one hand, the evaluation cost of industrialized sauce-marinated poultry meat sensory quality is high, and the evaluation index lacks unified standard, resulting in uneven data quality, which is difficult to support accurate modeling analysis; on the other hand, sauce-marinated poultry meat products of different raw material regions and auxiliary material sources have significant differences in flavor characteristics and consumer preference, and the existing methods lack consideration of these differences, resulting in insufficient generalization ability of the obtained model and algorithm. Therefore, a scientific, efficient and intelligent sauce-marinated poultry auxiliary material optimization and flavor directional design method and management system are needed, which can accurately depict the internal correlation between sauce-marinated poultry auxiliary material ingredients and sensory quality, so as to improve the consistency and stability of product quality, and realize intelligent control between auxiliary material and flavor sensory. SUMMARY

[0006] The purpose of the present application is to provide a multi-modal data fusion sauce-marinated product flavor directional design method and system, to solve the limitations of traditional sauce-marinated poultry sensory flavor quality consistency control and auxiliary material optimization intelligent management design and management problems, and to realize intelligent control between auxiliary material and flavor sensory.

[0007] To achieve the above-mentioned purpose, the present application provides a multi-modal data fusion sauce-marinated product flavor directional design method, comprising the following steps:

[0008] S1, constructing an auxiliary material database;

[0009] S2, using Apriori algorithm to mine the internal correlation of components in high-quality auxiliary materials;

[0010] S3, evaluating the sensory flavor of sauce-marinated products to obtain product index data set;

[0011] S4, establishing a multivariate mapping model between auxiliary material parameters and sensory quality based on Statsmodels Python library;

[0012] S5, combining consumer preference to establish a user score matrix;

[0013] S6, using genetic algorithm to optimize the auxiliary material component model, combining the multivariate mapping model obtained in S4 and the sensory weight obtained in S5 to solve, and obtaining the optimal quality auxiliary material components.

[0014] Preferably, in S1, the type of the auxiliary material database includes regional factors, auxiliary material types, auxiliary material ratios, auxiliary material addition methods, and consumer preference characteristics factors.

[0015] It is necessary to comprehensively and deeply master the related information such as the types and proportions of auxiliary materials added in the traditional sauce-marinated poultry meat making, so as to provide a solid foundation for subsequent determination of scientific, reasonable and traditional characteristic and market demand-oriented design of sauce-marinated poultry meat application formula and flavor. Through research and data collection, the unique style and auxiliary material components formed in the long-term development of traditional sauce-marinated poultry meat are collected, and the types, characteristics and adding methods of commonly used auxiliary materials are clarified, so as to better inherit and innovate in modernized processing, and make sauce-marinated poultry meat products more regional and market competitive.

[0016] Preferably, S2 is specifically:

[0017] S21, generating a candidate 1-item set: classifying and coding the raw material components in the auxiliary material database, and discretizing the adding number and proportion into quantitative values, taking the single auxiliary material item set as a candidate item set, and counting the frequency existing in the auxiliary material database;

[0018] S22, determining the frequent 1-item set: setting the minimum support, and screening out the candidate 1-item set with a frequency percentage not less than the minimum support;

[0019] S23, generating a candidate K-item set;

[0020] S24, pruning;

[0021] S25, determining the frequent K-item set;

[0022] S26, generating association rules from the frequent K-item set according to the set minimum confidence, realizing the mining of the internal association among the auxiliary material components of traditional sauce-marinated poultry meat, and providing a decision basis for subsequent optimization of auxiliary materials for flavor design.

[0023] Preferably, in S3, the 9-point scoring method is used to evaluate the product sensory flavor index in different auxiliary material components, and the product sensory flavor index includes texture, appearance and flavor.

[0024] These data can intuitively reflect the differences in the sensory quality of sauce-marinated poultry meat with different auxiliary material components, provide a scientific basis for subsequent optimization of auxiliary materials and directional design of flavor, and make the final sauce-marinated poultry meat product reach the best level in terms of sensory characteristics, so as to meet the needs of consumers for high-quality sauce-marinated products.

[0025] Among them, the sensory evaluation involves the selection of evaluators. Professional sensory evaluators should have acute sensory perception ability, and can be selected from food science students, personnel engaged in food sensory evaluation and experienced food lovers. At the same time, the evaluators should be healthy, without diseases in olfaction and taste, and have good expression ability and responsibility.

[0026] Sensory evaluation involves evaluation training. The training content includes: defining the definition and evaluation points of each sensory index, such as considering the uniformity, brightness and hue of the appearance index; focusing on the richness, purity and coordination of the aroma, and the balance and richness of the salty, sour, bitter and fresh tastes of the flavor index; the texture index includes the softness, elasticity and chewiness of the meat. During the training process, the evaluation standard and scoring scale need to be unified to ensure that the evaluation results between evaluators are consistent and comparable.

[0027] Sensory evaluation involves evaluation standards. Sensory evaluation is performed using quantitative descriptive analysis. The sensory evaluation results are recorded using a 9-point scale, and the scoring levels of 7-9, 4-6 and 1-3 are given according to "whether the texture and flavor are attractive, the appearance and organization state are tight, the elasticity is good, the juice is rich, and the meat aroma is pure". The same method is applicable to the evaluation of other indexes such as flavor, taste and texture. The texture, appearance, flavor and overall acceptability of each group of processed meat patties are scored systematically.

[0028] The above data can intuitively reflect the differences in sensory quality of different formula brine chickens, providing specific reference for subsequent formula optimization and flavor directional design. At the same time, through strict selection and systematic training of evaluators, as well as standardized evaluation process and environment setting, the consistency and reliability of the evaluation results are ensured. Based on the small scoring difference between different evaluators, the data also has high credibility and comparability. Further, through the arrangement and analysis of data, the comprehensive quality performance of high-quality auxiliary material components in each sensory index of sauce brine poultry meat products can be clearly analyzed.

[0029] Preferably, S4 is specifically:

[0030] S41, determining the research object and variable;

[0031] S42, data collection;

[0032] S43, data preprocessing by Z-score standardization;

[0033] S44, mapping model is established by using Statsmodels library.

[0034] Preferably, in S43, the rule of Z-score standardization is:

[0035]

[0036] Wherein, m i and are the i-th data value, μ and σ are the mean and standard deviation of the data population.

[0037] Preferably, in S5, the sensory weight ratio of different dimensions is obtained by constructing a user score matrix through consumer quality preference research and analytic hierarchy process.

[0038] The user score matrix is constructed in combination with consumer preference, a survey questionnaire with clear structure and clear questions is designed, the multi-dimensional preference degree of consumers for the flavor sensory indicators of sauce-marinated poultry meat is covered, and a scoring consumer group is set up to determine the real product experience by exchanging the flavor sensory characteristics of sauce-marinated products.

[0039] The judgment matrix is constructed and verified by analytic hierarchy process, and the weight of multi-dimensional sensory indicators is determined by calculating the characteristic vector and maximum eigenvalue of the judgment sentence. For subsequent optimization of sauce-marinated product auxiliary material composition, the actual performance of multi-dimensional flavor sensory indicators is also considered while ensuring consumer preference.

[0040] Preferably, S6 is specifically:

[0041] S61, auxiliary material coding: digitize the auxiliary material components, convert the amount of each auxiliary material into a gene form that can be processed by genetic algorithm, and obtain a chromosome;

[0042] S62, coding initialization: based on the internal correlation of S2, obtain the auxiliary materials corresponding to the initial chromosome;

[0043] S63, fitness evaluation: obtain the comprehensive value according to the multi-mapping model obtained in S4 and the weight in S5, take the comprehensive value as the objective function, and take the raw material cost as the constraint condition to calculate the fitness of each chromosome;

[0044] S64, select the chromosome by roulette selection method;

[0045] S65, crossover and mutation;

[0046] S66, iterative optimization: repeat S64-S65 multiple times, and when the maximum iteration number is reached or the fitness value does not improve continuously for 20 generations, terminate the iteration, and obtain the optimal auxiliary material components.

[0047] Preferably, in S65, the crossover operation principle is:

[0048] Suppose two formula individuals A=(a1, a2, …, a n ) and B=(b1, b2, …, b n ) are crossed at the defined crossover point k to generate offspring formula individuals C=(c1, c2, …, c n ) and D=(d1, d2, …, d n );

[0049] The mutation operation principle is: for formula individual A=(a1, a2, …, a n ), the component ai a' is obtained by performing a variation on a i a' = a i + Δf (t, y), wherein Δf (t, y) is a random function related to iteration number t and value range y.

[0050] Based on the optimization of sauce and marinade poultry meat product accessories and the flavor directional design level, the comprehensive weight value of multi-dimensional sensory indexes is taken as a target function, covering texture, appearance and flavor and the like, and the comprehensive optimization of these indexes represents that the sauce and marinade poultry meat product reaches the best state in flavor and quality. The raw material cost directly related to the production cost and market pricing of the product is set as a constraint condition, and needs to be reasonably controlled on the premise of guaranteeing quality. The formula optimization model based on sensory indexes is established by using a modified genetic algorithm. Therefore, the optimal flavor sensory and cost control of sauce and marinade poultry meat products can be realized on the premise of guaranteeing controllable and stable raw material cost, the product concept of directional flavor design is realized, the product quality and market competitiveness are improved, and a solid foundation is laid for the industrialized production and market promotion of sauce and marinade poultry meat products.

[0051] The application also provides a sauce and marinade product flavor directional design management system fusing multi-modal data, which comprises, which are connected in sequence, a sauce and marinade accessory acquisition module, an accessory component correlation module, a sensory data acquisition module, a quality mapping modeling module, a user score setting module and an accessory intelligent design module.

[0052] The sauce and marinade accessory acquisition module is used for constructing a database of accessory components and dosages of various regions and various industries.

[0053] The accessory component correlation module adopts a modified Apriori correlation algorithm to mine the correlation between qualitative and quantitative accessory components, and obtains a frequent item set.

[0054] The sensory data acquisition module is used for constructing a product sensory evaluation data set under different accessories.

[0055] The quality mapping modeling module is used for constructing a multi-element mapping model of accessory components and sensory indexes, and operates modeling based on a Python library of Statsmodels, so as to obtain the relationship between sauce and marinade meat accessory components and quantitative and different dimensional sensory evaluation attributes.

[0056] The user score setting module adopts a consumer quality preference research and an analytic hierarchy process to construct a user score matrix, and obtains a sensory weight ratio of different dimensions.

[0057] The accessory intelligent design module is used for outputting accessory parameters and accessory components.

[0058] Therefore, the sauce and marinade product flavor directional design method and system fusing multi-modal data have the following beneficial effects:

[0059] (1) The present application proposes a kind of fusion multimodal mining and optimization of sauce marinated poultry flavor directional design technical method and management system.The system uses the flavor improvement strategy of marinating auxiliary material optimization,It is characterized by integrating extensive data source collection,Multi-model architecture,Quantitative and qualitative mapping,Under the premise of guaranteeing the marinating application mode of traditional sauce marinated product,Realize the intelligent control of sauce marinated poultry product based on flavor sensory target and cost limit,Not only can guarantee the flavor consistency of product,Also can improve the fullness and coordination of sauce marinated poultry,So as to improve the product adaptation degree and satisfaction of consumer.

[0060] (2)The present application combines the technologies in multiple fields such as data mining,association analysis and optimization theory,fully considers the industry characteristics and actual production needs,solves the problems of traditional sauce marinated poultry sensory flavor quality consistency control limitations and auxiliary material optimization intelligent management design management,realizes the directional design of product flavor for special groups,forecasts the maximum reduction of process cost in advance,realizes the intelligent control between auxiliary material and flavor sensory,provides strong technical support for the quality improvement and personalized customization of traditional sauce marinated poultry products.

[0061] The technical solutions of the present application will be described in detail below with the help of drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 is a module architecture schematic diagram of a kind of fusion multimodal data of the present application sauce marinated product flavor directional design system embodiment;

[0063] Figure 2 is a flow chart of a kind of fusion multimodal data of the present application sauce marinated product flavor directional design method embodiment;

[0064] Figure 3 is the Apriori algorithm association logic diagram of S2 of a kind of fusion multimodal data of the present application sauce marinated product flavor directional design method embodiment;

[0065] Figure 4 is the genetic algorithm optimization logic diagram of S6 of a kind of fusion multimodal data of the present application sauce marinated product flavor directional design method embodiment. DETAILED DESCRIPTION

[0066] The technical solutions of the present application will be described in detail below with the help of drawings and examples.

[0067] Unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meaning understood by persons with ordinary skills in the art to which the present application belongs.

[0068] Example one

[0069] AsFigure 1 As shown, a halogen product flavor oriented design management system fusing multi-modal data includes a halogen auxiliary material collection module 01, an auxiliary material component association module 02, a sensory data acquisition module 03, a quality mapping modeling module 04, a user score formulation module 05, and an auxiliary material intelligent design module 06 connected in turn.

[0070] Specifically, the principles of the halogen auxiliary material collection module 01 are as follows:

[0071] 1) Raw material research: Detailed records of the halogen process of traditional sauce halogen products, including the preparation of halogen soup, halogen time, temperature control, etc., as well as the types, brands and approximate amounts of auxiliary materials. Adopt standardized production process, for example, the amount of a certain auxiliary material has a relatively fixed range according to the production scale. In a more specific scheme, the scheme needs to understand the traditional individual's production process and auxiliary material components, and the conventional individual has the traditional experience and secret recipe.

[0072] 2) Raw material types: According to the research results, determine the common halogen material types, such as star anise, cassia bark, linaloe, Sichuan pepper, fennel, etc. At the same time, pay attention to collect some auxiliary materials with regional traditional characteristics and unique flavor effects in sauce halogen poultry meat.

[0073] The principles of the auxiliary material component association module 02 are as follows:

[0074] For the halogen auxiliary materials collected by the 01 module, the Apriori association rule mining algorithm is used. The ingredients in the formula are coded to build a transaction data set. Through algorithm mining, the association relationship between different ingredients is found out, and the frequently occurring ingredient combination mode is found out, realizing the depth analysis of the auxiliary materials determined by the 01 module, and providing data support for subsequent model establishment. 02 excavates potential value information from traditional auxiliary materials, aiming to make a key transition from experienced auxiliary materials to scientific optimization.

[0075] The principles of the sensory data acquisition module 03 are as follows:

[0076] According to the application of the 01 module, halogen auxiliary materials are made into multiple batches of sauce halogen meat product samples. Invite a professional sensory evaluation team, including food professionals, experienced consumers, etc., to evaluate the texture, appearance and flavor of sauce halogen products. Further, the 03 module can be supplemented by the ingredient association data excavated by the 02 module, and together provide the basis for model construction. It provides direction guidance for optimizing halogen auxiliary materials from the perspective of actual sensory performance of products.

[0077] The principles of the quality mapping modeling module 04 are as follows:

[0078] The multiple linear regression analysis method is used, the raw material components are taken as independent variables, and the multi-dimensional sensory indexes are taken as dependent variables. Through data fitting, a regression mapping model between the components of the sauce and marinade product auxiliary materials and the sensory indexes is established. The model is a bridge connecting the auxiliary materials and the sensory indexes, and can quantitatively describe how the changes of the raw material components affect the sensory indexes, thereby providing a quantitative tool for subsequent formula optimization. The 04 module can realize the conversion from data to model, and provide theoretical guidance for the actual flavor directional design operation.

[0079] The principles of the user score development module 05 are as follows:

[0080] Market research is carried out to collect consumer preference data for different sensory characteristics of sauce and marinade products obtained by the above-mentioned multiple auxiliary materials. The relative importance of consumers to the appearance, flavor, texture and other sensory indexes can be understood through questionnaire survey, interview and other methods. According to the research results, each sensory index is given a corresponding weight. Combined with the sensory index data determined by the 03 module, a user score matrix is established. Each sensory latitude corresponds to a comprehensive weight in the matrix, which reflects the performance of the sensory latitude in meeting consumer preferences. Through this scoring matrix, the advantages and disadvantages of different senses can be directly compared, providing decision-making basis for selecting the best auxiliary materials based on sensory indexes. It is the key link to combine all the previous steps with market demand, and to ensure that the final product can meet the expectations of consumers.

[0081] The principles of the auxiliary material intelligent design module 06 are as follows:

[0082] Based on the sensory multiple mapping model established by the 04 module and the sensory weight score matrix of the 05 module, the 06 module sets the optimization objective function with the goal of meeting the expectations of consumers for sensory indexes. Considering the raw material cost as a constraint condition, the improved genetic algorithm is used to optimize the collected sauce and marinade meat product auxiliary material components and usage. Through continuous iterative calculation, the auxiliary material ratio that can maximize the objective function value under the condition of meeting the constraint condition is found. Specifically, the 06 module is based on the establishment of the aforementioned model and method, and uses optimized components to realize scientific adjustment of auxiliary materials to achieve better sensory effect and economic benefit, which is the practical application and deepening of the previous steps.

[0083] Example Two

[0084] As shown in Figure 2 A flavor directional design method for marinated products by fusing multi-modal data includes the following steps:

[0085] S1, construct an auxiliary material database:

[0086] The auxiliary material components and the amount of traditional sauce marinated poultry products are investigated in enterprises and individual households, and the initial application of auxiliary materials is collected. Specifically, common marinating auxiliaries such as star anise, cassia bark, linden leaf, Sichuan pepper, fennel, etc. are collected, combined with actual investigation and traditional process, and the application of auxiliary materials is preliminarily determined. For example, star anise [X1] grams, cassia bark [X2] grams, linden leaf [X3] grams, etc. Here [X1], [X2], [X3] specific values can be adjusted according to subsequent steps.

[0087] S2, using Apriori algorithm to mine the internal correlation of components in high-quality auxiliary materials, that is, analyzing the large amount of auxiliary material composition and amount data collected in S1, solving and obtaining the potential correlation between the initial application of auxiliary materials, such as if the frequency of the simultaneous appearance of star anise and cassia bark is high, this association rule can be determined to provide basis for formula optimization. For example Figure 3 As shown in the following steps:

[0088] S21, generating candidate 1-item set: taking a single auxiliary material component as a single candidate 1-item set, and counting the frequency of appearance in the auxiliary material database obtained in step S1.

[0089] S22, determine frequent 1-item set: set the minimum support (the conventional threshold is set to 30%), screen out the candidate 1-item set with a frequency percentage not less than the minimum support, if less than it can be discarded, the screened structure can be a frequent 1-item set.

[0090] Among them, the calculation formula of support is as follows:

[0091] Support (Support):

[0092] Among them, X is a certain formula component; transaction is the composition of the formula.

[0093] S23, generating candidate K-item set: generating candidate K-item set by merging operation of two frequent (K-1)-item sets, (K is a random number);

[0094] S24, pruning: pruning operation is performed on the generated candidate K-item set, if a (K-1)-item set subset of a candidate K-item set is not frequent, it means that this candidate K-item set is also not possible to be frequent, then pruning and deleting is selected; In addition, according to the set minimum support, further pruning and deleting infrequent item set;

[0095] S25, determine frequent K-item set; repeat steps S23-S24 until no new frequent K-item set is generated to stop;

[0096] S26, generating association rules from the frequent K-item set according to the set minimum confidence (the conventional confidence threshold is 60%). If the calculated confidence is not less than the minimum confidence threshold, the association rules can be retained.

[0097] The confidence is calculated according to the following formula:

[0098] Confidence:

[0099] Wherein, X→Y is the association degree of formula ingredient X and formula ingredient Y.

[0100] The steps of the example use the Apriori algorithm to mine frequent item sets and association rules from various traditional sauce-marinated poultry products and accessories, providing a solution basis for subsequent mapping model establishment and accessory optimization.

[0101] S3, evaluating the sensory flavor of marinated products to obtain a product index data set. The present embodiment uses a 9-point rating method to quantify and score each sensory index. At the same time, relevant descriptive information during the evaluation process is recorded. All evaluation results are summarized to form a product index data set. Specifically, professional sensory evaluators are invited to conduct sensory evaluation processes according to the modified sensory evaluation form (as shown in Table 1), sensory evaluation of sauce-marinated poultry products with different accessory compositions is conducted, and a 9-point rating method is used to obtain a product index data set.

[0102] Table 1 Sensory evaluation form

[0103]

[0104] In the example of this step, sauce-marinated poultry samples with different accessory compositions are randomly numbered to avoid the influence of evaluation results by evaluators due to numbering order or other subjective factors. The numbering can be in the form of numbers, letters, or a combination of both, such as A1, B2, etc. Each sample should be presented under the same conditions, such as using the same dinner plate, placement, etc. Evaluators need to follow standardized evaluation procedures and set up a uniform sensory evaluation room environment to ensure the consistency and reliability of the evaluation results, so that the scoring difference between different evaluators is small, and the data has high credibility and comparability.

[0105] The product index data set can be used as an important reference for establishing quality standards for sauce-marinated poultry products, which helps enterprises to carry out standardized production and quality control, and ensures the stability and consistency of product quality. The product index data set provides a scientific basis for the optimization of sauce-marinated poultry accessories. According to the performance of each accessory in the sensory index, the types and amounts of accessories are adjusted accordingly, thereby improving the overall sensory quality of sauce-marinated poultry, and achieving the best balance in texture, appearance, flavor, etc.

[0106] S4, a multivariate mapping model between the auxiliary material parameters and the sensory quality was established by using the multivariate regression analysis method based on the Python library of Statsmodels, and the specific steps were as follows:

[0107] S41, determining the research object and variable: the product index data set obtained in step S3 was used as the dependent variable (Y), and the independent variable (X) can be set as the amount of star anise (unit: gram), the amount of cassia (unit: gram), the amount of rosemary (unit: gram), the amount of Sichuan pepper (unit: gram) and other auxiliary material compositions.

[0108] S42, data collection: the product index data set obtained in step S3 was used.

[0109] S43, data preprocessing: Z-score standardization was used to convert the data of the independent variable and the dependent variable into standard normal distribution data with a mean of 0 and a standard deviation of 1.

[0110] Among them, the rules of Z-score standardization are as follows:

[0111]

[0112] Among them, m i and are the i-th data value, μ and σ are the mean and standard deviation of the data population.

[0113] S44, establishing a mapping model: the Statsmodels library was used to calculate the coefficients of the standard model function according to the input data, including the coefficient of the constant term and the coefficient of the independent variable.

[0114] Among them, the main function functions are shown in Table 2:

[0115] Table 2 Function functions and their functions for constructing mapping models

[0116]

[0117]

[0118] The generated standard model function is as follows:

[0119] Y = β0 + β1x1 + β2x2 + β3x3 + … + β n x n + σ

[0120] Among them, Y: product index data set, i.e. sensory index score (dependent variable), X: formula component amount (independent variable), β0: intercept term, β1, β2, β3…β n : regression coefficient, σ: error term (sensory index score when no formula is adopted).

[0121] S45, Model evaluation: use the coefficient of determination (R 2 ), to evaluate the goodness of fit of the model. R 2 The closer to 1, the better the model fitting effect on the data, that is, the higher the degree to which the independent variable auxiliary component can explain the change of the dependent variable sensory index. At the same time, the uncertainty of the beta coefficient is quantified by using the confidence interval, which helps to judge the statistical significance of the independent variable.

[0122] In this step example, the multi-dimensional sensory index data of sauce-marinated poultry meat under the corresponding auxiliary component is predicted, which provides data support and theoretical basis for subsequent auxiliary optimization.

[0123] S5, combined with the establishment of consumer preference, as follows:

[0124] S51, consumer preference survey: design a questionnaire with clear structure and clear questions, covering the preference degree of consumers for the texture, appearance and flavor of sauce-marinated poultry meat products.

[0125] S52, establish user score matrix: require expert evaluation group, and first construct judgment matrix based on analytic hierarchy process. For multi-dimensional sensory index, determine the corresponding position assignment (1-9 scale method, 1 indicates equal importance, 3 indicates slightly important, 5 indicates obviously important, 7 indicates strongly important, and 9 indicates extremely important) in the judgment matrix according to the relative importance scale table (as shown in Table 3). Calculate the maximum eigenvalue λ max and its corresponding eigenvector W of the judgment matrix A, and obtain the weight vector after normalizing the eigenvector.

[0126] Table 3 relative importance 1-9 scale table

[0127] Scale value Meaning 1 Both indicators are equally important 3 The former is slightly more important than the latter 5 The former is significantly more important than the latter 7 The former is strongly more important than the latter 9 The former is extremely more important than the latter Reciprocal The latter is more important than the former (e.g. 1 / 5 indicates that the latter is significantly more important)

[0128] S53, consistency check: calculate the consistency index where n is the order of the judgment matrix A; the consistency ratio is the random consistency index RI, and when the consistency ratio CR is less than 0.1, it is considered that the judgment matrix A has satisfactory consistency, that is, the obtained index weight can be used.

[0129] In this step example, the weight vector obtained by solving comprehensively considers the actual performance of each index, providing comprehensive basis for subsequent targeted flavor-oriented design of auxiliary optimization, and ensuring that the product obtains higher recognition and competitiveness.

[0130] S6, optimize the auxiliary material component model using genetic algorithm, combine the multi-mapping model obtained in S4 and the sensory weight obtained in S5 to solve, take the sensory index comprehensive value (according to the weight vector obtained in step S5) as the objective function, the sensory index covers the aforementioned flavor, texture and appearance, etc., and the comprehensive optimization of these indexes represents that the sauce-marinated poultry meat product reaches the best state in flavor and quality. Because the raw material cost is directly related to the production cost and market pricing of the product, it is necessary to reasonably control under the premise of ensuring quality, set the raw material cost as a constraint condition, and finally obtain the auxiliary material component with optimal quality. As shown in the following steps: Figure 4

[0131] S61, auxiliary material coding: digitize the auxiliary material components and convert the amount of each auxiliary material into a gene form that can be processed by genetic algorithm, such as [amount of star anise, amount of cassia bark, amount of rosemary, amount of Sichuan pepper, amount of fennel], this group of real numbers constitutes a "chromosome", representing a complete sauce-marinated poultry meat auxiliary material;

[0132] S62, coding initialization: each initial "chromosome" corresponds to an auxiliary material based on the collection in step S1 and the association rules obtained in step S2;

[0133] S63, fitness evaluation: according to the objective function and the constraint condition, the fitness of each "chromosome" (auxiliary material composition) is calculated, wherein the fitness function is used to evaluate the degree of excellence of each individual (i.e. formula composition), for example, the fitness function Fitness = f (comprehensive sensory score) can be defined according to the predicted sensory index score of the regression model.

[0134] The objective function is guided by the comprehensive value of the sensory index, through the multi-mapping model established before (the result of step S4), the multi-dimensional sensory index data is predicted, and then the comprehensive value is obtained through the sensory weight of S5; then the cost of each raw material is calculated according to the market price, if the cost exceeds the preset upper limit, the fitness is reduced; if the cost exceeds the upper limit, the fitness is the sensory index score minus a certain proportion (usually set as twice the excess cost) of the excess cost.

[0135] Specifically, through this comprehensive evaluation method, the fitness value of each formula is obtained, the higher the fitness, the closer the formula is to the optimal solution.

[0136] S64, roulette wheel selection: adopt roulette wheel selection method, according to the fitness, the higher the fitness of the "chromosome" in the population, the larger the sector area it occupies, and the greater the probability of being selected.

[0137] ​Specifically, roulette wheel rotation is simulated by generating a random number between 0 (inclusive) and 1 (exclusive). In roulette selection, the individual with the highest fitness value is not necessarily selected, but is randomly selected according to probability, which ensures the diversity of the population and avoids premature convergence to a local optimal solution.

[0138] wherein the roulette selection method principle: the probability of individual i being selected wherein Fitness i is the fitness value of individual i, and N is the population (formula group) size.

[0139] S65, crossover and mutation: single-point crossover is used for the composition of the auxiliary material, a crossover point is randomly selected, and mutation is performed on multiple auxiliary material compositions, the nth auxiliary material component is randomly selected for mutation, and a basis is provided for generating a better new individual.

[0140] Crossover operation principle: assume that two formula individuals A = (a1, a2, …, a n ) and B = (b1, b2, …, b n ) are crossed at a defined crossover point k to generate offspring formula individuals C = (c1, c2, …, c n ) and D = (d1, d2, …, d n ).

[0141] Mutation operation principle: for formula individual A = (a1, a2, …, a n ), the component a i is mutated to a′ i = a i + Δf(t, y), where Δf(t, y) is a random function related to the iteration number t and the value range y.

[0142] S66, iterative optimization: S64-S65 is repeated multiple times, and a new "chromosome" (auxiliary material composition) is generated for each generation of population. In each iteration, the optimal "chromosome" (auxiliary material with the highest fitness) and its fitness value in the current population are recorded.

[0143] wherein the termination principle is to set the maximum number of iterations to 500, and when the maximum number of iterations is reached or the fitness value does not significantly improve after 20 consecutive generations, the iteration is terminated. After multiple iterations, the "chromosome" (auxiliary material component) in the sauce and marinated poultry meat gradually approaches the optimal solution (the best comprehensive sensory value), and finally the sauce and marinated poultry meat auxiliary material that meets the constraint condition and has the optimal sensory index is obtained.

[0144] In the example of the present step, the improved genetic algorithm is used to establish an excipient optimization model based on sensory indicators. Under the premise of ensuring the cost of raw materials, the optimal sensory indicators of sauce-marinated poultry meat are obtained, thereby realizing the comprehensive product flavor design concept based on consumer preferences and improving the product quality and market competitiveness.

[0145] Therefore, the present application adopts the above-mentioned flavor directional design method and system for marinated products based on fusion of multi-modal data, aiming to solve the problems of consistency control limitation of traditional sauce-marinated poultry meat sensory flavor quality and intelligent management design management of excipient optimization, and realize intelligent control between excipient-flavor sensory.

[0146] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit them. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can still be modified or replaced by equivalents, and these modifications or equivalent replacements should not make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for designing flavor orientation of marinated products by integrating multimodal data, characterized in that: The following steps are included: S1. Build auxiliary material database; S2. Use Apriori algorithm to mine the intrinsic correlation of components in high-quality excipients; S3. Evaluate the sensory flavor of marinated products and obtain product index data sets; S4. Establish a multivariate mapping model between excipient parameters and sensory quality based on the Statsmodels Python library; S5. Establish a user rating matrix based on consumption preferences; S6. Use genetic algorithm to optimize the auxiliary material component model, combine the multivariate mapping model obtained in S4 and the sensory weight obtained in S5 to solve and obtain the auxiliary material component with the best quality.

2. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 1, characterized in that: In S1, the types of the auxiliary material database include regional factors, auxiliary material types, auxiliary material ratios, auxiliary material adding methods, and consumer preference characteristics.

3. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 1, characterized in that: S2 is specifically: S21. Generate candidate 1-item sets: Classify and encode the raw material components in the excipient database, discretize the addition quantity and ratio into quantitative values, and take the individual excipient item set as a candidate item set, and count the frequency of its existence in the excipient database; S22. Determine frequent 1-item sets: set the minimum support and select candidate 1-item sets whose frequency percentage is not less than the minimum support; S23, generating candidate K-item sets; S24, pruning; S25. Determine frequent K-item sets; S26. Generate association rules from the frequent K-item set according to the set minimum confidence.

4. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 1, characterized in that: In S3, a 9-point scoring method was used to evaluate the sensory flavor indicators of products in different auxiliary material components. The sensory flavor indicators of products included texture, appearance and flavor.

5. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 1, characterized in that: S4 is specifically: S41. Determine the research subjects and variables; S42, data collection; S43, data preprocessing using Z-score standardization; S44. Use the Statsmodels library to build a mapping model.

6. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 5, characterized in that: In S43, the Z-score normalization rule is: Among them, m i and are the i-th data value, μ and σ are the mean and standard deviation of the data population.

7. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 1, characterized in that: In S5, the sensory weight ratios of different dimensions are obtained by constructing a user rating matrix through consumer quality preference surveys and hierarchical analysis.

8. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 1, characterized in that: S6 specifically: S61. Excipient coding: digitally encode the excipient components and convert the amount of each excipient into a gene form that can be processed by the genetic algorithm to obtain a chromosome; S62, encoding initialization: based on the intrinsic association of S2, obtain the auxiliary materials corresponding to the initial chromosome; S63, fitness evaluation: according to the multivariate mapping model obtained in S4 and the weights in S5, a comprehensive value is obtained. The comprehensive value is used as the objective function and the raw material cost is used as the constraint condition to calculate the fitness of each chromosome; S64, chromosome selection using roulette wheel selection method; S65, Crossover and Mutation; S66. Iterative optimization: Repeat S64-S65 multiple times. When the maximum number of iterations is reached or the fitness value does not improve after 20 consecutive generations, terminate the iteration to obtain the auxiliary material component with the best quality.

9. The method for designing flavor orientation of marinated products by integrating multimodal data according to claim 8, characterized in that: In S65, the crossover operation principle is: Assume that there are two formula individuals A=(a1, a2,…, a n ) and B=(b1,b2,…,b n ) is crossed with the defined crossover point k to generate the offspring recipe individual C=(c1,c2,…,c n ) and D=(d1,d2,…,d n ); The variation operation principle: for the formula individual A=(a1, a2,…, a n ), for component a i Mutate into a′ i =a i +Δf(t,y), where Δf(t,y) is a random function related to the number of iterations t and the value range y.

10. A management system using the method for designing flavor orientation of marinated products by integrating multimodal data according to any one of claims 1 to 9, characterized in that: It includes a marinated auxiliary material collection module, an auxiliary material component association module, a sensory data acquisition module, a quality mapping modeling module, a user rating formulation module and an auxiliary material intelligent design module, which are connected in sequence; The braising auxiliary material collection module is used to systematically construct a database of auxiliary material components and usage in various regions and industries; The auxiliary material component association module uses the improved Apriori association algorithm to mine the association between the qualitative and quantitative aspects of auxiliary material components to obtain frequent item sets; The sensory data acquisition module is used to construct a product sensory evaluation data set under different excipients; The quality mapping modeling module is used to construct a multivariate mapping model between auxiliary material components and sensory indicators, and to perform operational modeling based on the Python library of Statsmodels to obtain the relationship between the auxiliary material components of braised meat and the quantitative and different dimensional sensory evaluation attributes; The user rating formulation module uses consumer quality preference research and hierarchical analysis method to construct a user rating matrix to obtain the sensory weight ratio of different dimensions; The auxiliary material intelligent design module is used to output auxiliary material parameters and auxiliary material components.

Images