Method for establishing cheese mature period evaluation based on machine learning and GC-MS (Gas Chromatography-Mass Spectrometer)

Abstract

The invention discloses a cheese mature period evaluation method based on machine learning and GC-MS. The cheese mature period evaluation method relates to the technical field of intelligent management of food production, and utilizes cheese flavor data measured by GC-MS and combines a classification model and feature engineering of machine learning to evaluate the mature period of cheese. A plurality of cheese characteristic flavor substances from ketone, acid and lactone are screened out to replace a large number of flavor substances to judge the mature period of cheese, the flavor characteristics of cheese in different mature periods are clarified through a decision tree, and then the flavor characteristics are utilized to judge the mature period of cheese in an unknown mature period. Automatic classification and flavor characteristic quantitative analysis in the cheese mature period are realized, and formula and process optimization can be supported, so that the cheese can meet the expected mature period.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for food production, and more specifically to a method for establishing a cheese ripening period evaluation model based on machine learning and GC-MS. Background Technology

[0002] Cheese, as a nutritious and uniquely flavored food, has a complex and diverse composition of flavor compounds, which play a crucial role in its quality and consumer acceptance. Traditional methods for evaluating cheese maturity mainly rely on manual sensory assessment or single chemical indicator detection, which suffers from problems such as strong subjectivity, low efficiency, inability to comprehensively reflect the contribution of multiple substances to the overall flavor, and difficulty in comprehensively and accurately identifying key characteristic flavor compounds.

[0003] In existing technologies, gas chromatography-mass spectrometry (GC-MS) can detect hundreds of volatile flavor compounds in cheese. However, directly using full-spectrum data for modeling leads to the curse of dimensionality and lacks a screening mechanism for key flavor markers. Furthermore, while existing machine learning methods (such as PCA and PLS) can reduce dimensionality, they lack the ability to capture nonlinear relationships, have insufficient interpretability of feature importance, and have not designed efficient feature screening processes specifically for the characteristics of the food industry, resulting in high data acquisition costs.

[0004] Therefore, how to accurately and efficiently evaluate the maturity period of cheese is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method for evaluating cheese maturity based on machine learning and GC-MS. By combining machine learning models with gas chromatography-mass spectrometry (GC-MS) technology, key flavor substances are screened through feature engineering, realizing automated classification of cheese maturity and quantitative analysis of flavor characteristics. It can also support the optimization of formulas and processes to meet the desired maturity.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for evaluating cheese maturity based on machine learning and GC-MS includes the following steps:

[0008] Step 1: Collect flavor compound data from cheese samples at different maturity stages using gas chromatography-mass spectrometry (GC-MS) and construct a flavor compound dataset; the flavor compound data includes the types and contents of flavor compounds.

[0009] Step 2: Perform augmentation preprocessing on the flavor compound dataset to obtain the training dataset;

[0010] Step 3: Build a machine learning classification model, and use the training dataset to train and optimize it to obtain a cheese maturity evaluation model;

[0011] Step 4: Optimize the types of input flavor compounds for the cheese ripening period evaluation model using the feature selection method based on the training dataset, and determine the characteristic flavor compounds; obtain a cheese ripening period evaluation model with high accuracy and few input flavor compounds;

[0012] Step 5: The optimized cheese ripening evaluation model outputs a decision tree, displaying the characteristic flavor compounds and their contents corresponding to cheeses at different ripening stages;

[0013] Step 6: Collect characteristic flavor substance data of the cheese to be evaluated and input them into the optimized cheese maturity evaluation model to obtain the cheese maturity evaluation results.

[0014] Preferably, the flavor compound dataset includes the types and corresponding contents of flavor compounds in cheeses at different maturity stages; the flavor compound dataset is processed using synthetic minority oversampling technology to obtain the training dataset.

[0015] Preferably, the training dataset is divided into a training set and a test set in a 7:3 ratio. The training set is used to train the model, and the test set is used to test the trained model.

[0016] Preferably, in step 3, multiple machine learning classification models are constructed and trained separately. The machine learning classification model with the highest accuracy and best performance is selected as the cheese ripening period evaluation model, that is, the best model is selected based on model performance; the model parameters are optimized using parameter tuning methods. The model with the best performance is selected based on model performance indicators.

[0017] Preferably, the machine learning classification models include, but are not limited to, Extreme Gradient Boosting (XGboost), K-Nearest Neighbor (KNN), Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Deep Learning (DL). Different tuning parameters are set for different models, and the best-performing model is obtained by combining 5-fold cross-validation, F1 score, precision, and recall as the cheese ripening period evaluation model. The Random Forest model is the optimal model.

[0018] Preferably, parameter tuning methods include, but are not limited to, grid search, random search, Bayesian optimization, and automated parameter tuning.

[0019] Preferably, feature selection methods include, but are not limited to, SelectKBest, Recursive Feature Elimination (RFE), PCA, LASSO, L1 regularization, etc., with the recursive feature elimination method being the optimal choice.

[0020] Preferably, the stopping condition for model optimization iteration in step 4 is set to the point where the amount of remaining flavor compounds reaches the expected amount, or the performance index of the cheese maturity evaluation model on the test set of the training dataset no longer improves; performance indexes include accuracy or F1 score, etc.

[0021] Preferably, when using the recursive feature removal method for optimization, all kinds of flavor substances and their contents in the training dataset are input into the cheese ripening period evaluation model, the importance scores of each flavor substance are calculated, and the flavor substances with the lowest scores are removed according to their importance. The remaining flavor substances and their contents are used as input to retrain the model. This process is repeated until the iteration stops, and the flavor substances finally selected are used as feature flavor substances.

[0022] Preferably, the cheese maturity evaluation model performs SHAP interaction analysis on the characteristic flavor compounds and their contents at different maturity stages, calculates the SHAP interaction value, and generates the contribution of characteristic flavor compounds of cheese at different maturity stages.

[0023] Preferably, the cheese maturity evaluation model uses a decision tree algorithm to generate a decision tree based on information gain or Gini coefficient, and uses the shortest path of each cheese maturity stage in the decision tree to determine the characteristic flavor substances and content of cheeses at different maturity stages.

[0024] Preferably, the process also includes step 7, which generates a cheese processing adjustment strategy based on the maturity evaluation results. This includes adjusting the maturity time, maturity temperature, type and content of added enzymes, and amino acid content to meet the desired maturity period, thereby producing high-quality cheese with a short maturity period and improving cheese quality.

[0025] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method for evaluating cheeses at different maturity stages based on machine learning and GC-MS. Through dynamic feature screening and automated parameter optimization, it screens key substances that significantly contribute to cheese flavor from complex cheese flavor compounds, eliminates redundant and unimportant flavor compounds, and achieves high-precision classification and visual analysis of flavor mechanisms. This provides a scientific basis for production process optimization and new product development, while solving the problems of data redundancy, poor model generalization, and opaque mechanisms in traditional methods. This invention is data-driven and highly automated. Compared with traditional manual screening methods, it greatly improves screening efficiency, can quickly screen characteristic flavor compounds from a large number of flavor compounds, and is highly efficient. It is not only applicable to specific types or origins of cheese, but also has good applicability to the screening of flavor compounds in different types of cheese and other foods. It provides strong technical support for quality control and product development in the food industry and has universal applicability. Attached Figure Description

[0026] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0027] Figure 1 This invention provides a schematic diagram of a method for evaluating cheeses at different maturity stages based on machine learning and GC-MS.

[0028] Figure 2 A schematic diagram showing the accuracy of different classification models provided by this invention after training.

[0029] Figure 3 This is a schematic diagram of the ROC curves after training of different classification models provided by the present invention;

[0030] Figure 4 SHAP interaction diagram of characteristic flavor compounds of cheese during the 3-month maturation period provided by the present invention;

[0031] Figure 5 SHAP interaction diagram of characteristic flavor compounds of cheese during the 6-month maturation period provided by the present invention;

[0032] Figure 6 SHAP interaction diagram of characteristic flavor compounds of cheese during the 9-month maturation period provided by the present invention;

[0033] Figure 7 SHAP interaction diagram of characteristic flavor compounds of cheese during the 10-month maturation period provided by the present invention;

[0034] Figure 8 SHAP interaction diagram of characteristic flavor compounds of cheese during the 12-month maturation period provided by the present invention;

[0035] Figure 9 This is a partial decision tree diagram for determining cheeses at different maturity stages, provided by the present invention. Detailed Implementation

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] This invention discloses a method for evaluating cheese maturity based on machine learning and GC-MS, such as... Figure 1 As shown, it includes the following steps:

[0038] S1: Collect flavor compound data from cheese samples at different maturity stages using gas chromatography-mass spectrometry (GC-MS) and construct a flavor compound dataset; the flavor compound data includes the types and contents of flavor compounds.

[0039] S2: Enhance the flavor compound dataset by preprocessing to obtain the training dataset;

[0040] S3: Build a machine learning classification model, and use the training dataset for training and optimization to obtain a cheese maturity evaluation model;

[0041] S4: Optimize the types of input flavor compounds for the cheese ripening period evaluation model using the feature selection method based on the training dataset, and determine the characteristic flavor compounds;

[0042] S5: Output the optimized cheese ripening evaluation model as a decision tree, showing the characteristic flavor compounds and their contents of cheeses at different ripening stages;

[0043] S6: Collect characteristic flavor substance data of the cheese to be evaluated and input them into the optimized cheese maturity evaluation model to obtain the cheese maturity evaluation results.

[0044] Furthermore, the maturation periods for the collected flavor compound data included 3 months, 6 months, 9 months, 10 months, and 12 months.

[0045] Furthermore, a synthetic minority oversampling technique was used to process the flavor substance dataset to obtain a training dataset.

[0046] Furthermore, the training dataset is divided into a training set and a test set in a 7:3 ratio. The training set is used to train the model, and the test set is used to test the trained model.

[0047] Furthermore, in S3, multiple classification models are constructed and trained separately. The machine learning classification model with the highest accuracy and best performance is selected as the best model. The model parameters of the best model are optimized using a parameter tuning algorithm to obtain the cheese maturity evaluation model.

[0048] Furthermore, machine learning classification models include Extreme Gradient Boosting (XGboost), K-Nearest Neighbor (KNN), Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Deep Learning (DL). Different tuning parameters are set for different models, and the best-performing model is obtained by combining 5-fold cross-validation, F1 score, precision, and recall as the cheese ripening period evaluation model.

[0049] Furthermore, parameter tuning algorithms include methods such as grid search, random search, Bayesian optimization, and automated parameter tuning.

[0050] Furthermore, feature selection methods include SelectKBest, Recursive Feature Elimination (RFE), PCA, LASSO, L1 regularization, etc., with the optimal choice being the recursive feature elimination method.

[0051] Furthermore, the stopping condition for model training is set to the point where the amount of remaining flavor compounds reaches the expected amount, or the model's performance metrics on the test set no longer improve; performance metrics include accuracy and F1 score, etc.

[0052] Furthermore, when using the RFE method for optimization, all types of flavor compounds and their corresponding contents in the dataset are input into the cheese ripening period evaluation model. The importance scores of each flavor compound are calculated, and flavor compounds are selected based on their importance scores. The corresponding contents are then iteratively input into the cheese ripening period evaluation model until the iteration stops. The flavor compounds selected at the end are taken as the characteristic flavor compounds.

[0053] Furthermore, the cheese maturity evaluation model conducts SHAP interaction analysis on the characteristic flavor compounds and their contents at different maturity stages, calculates SHAP interaction values, and generates the contribution of characteristic flavor compounds of cheese at different maturity stages.

[0054] Furthermore, the cheese maturity evaluation model uses a decision tree algorithm to generate a decision tree based on information gain or Gini coefficient, and uses the shortest path of each cheese maturity stage in the decision tree to determine the characteristic flavor substances and content of cheeses at different maturity stages.

[0055] Furthermore, using a random forest combined with a recursive feature deletion method, 14 characteristic flavor compounds of cheese were screened, including ketones, acids, and lactones. The ketones included 2-heptanone, 2-nonanone, 2-undecanone, and acetoin; the acids included acetic acid, butyric acid, pentanoic acid, hexanoic acid, octanoic acid, nonanoic acid, and n-decanoic acid; and the lactones included γ-dodecalactone, δ-dodecalactone, and δ-decanolactone.

[0056] Furthermore, S7 includes generating cheese processing adjustment strategies based on the maturity evaluation results. This includes adjusting maturity time, maturity temperature, the type and content of added enzymes, and the amino acid content to meet the desired maturity period, producing high-quality cheese with a short maturity period and improving cheese quality. The volatile flavor characteristics of cheese with unknown maturity periods or fast-matured cheese are compared with the volatile flavor characteristics of cheeses with different maturity periods derived from the model to determine the cheese's maturity period and formulate cheese processing adjustment strategies.

[0057] On the other hand, in one specific embodiment, the method for evaluating cheese maturity based on machine learning and GC-MS is experimentally verified, including the following steps:

[0058] (1) Data collection: Data on the flavor of cheddar cheese were collected from literature and experiments. In order to increase the dataset size and improve the model performance, data were selected from flavor compounds measured by headspace solid phase extraction-GC-MS. The data were limited to full-fat cheddar cheese made with milk and rennet, commercially available cheddar cheese aged for 3 to 12 months, and the volatile flavor compound content was expressed in μg / kg.

[0059] The method for determining volatile flavor by GC-MS is as follows: 3g of ground cheese is placed in a 20mL headspace vial, and 3μL of 2-methyl-3-heptanone (mass concentration 0.816mg / mL) is added as an internal standard. The mixture is equilibrated at 600r / min and 60℃ for 30min. Then, the aged extraction head is inserted into the headspace vial for extraction, and the adsorption time is 30min. The extraction head is then inserted into the gas chromatography vaporization chamber and desorbed at 250℃ for 5min. GC-MS determination was performed using a gas chromatography-triple quadrupole mass spectrometer (Agilent 7000D GC 8890-MS / MS) equipped with an HP-Innowax column (Agilent 19091N-133l, 30m×250μm×0.25μm). Chromatographic conditions: 35℃ for 3 min, then increased to 120℃ at 4℃ / min, then to 190℃ at 5℃ / min, and finally to 230℃ at 10℃ / min, held for 6 min, splitless injection, purge flow rate to the split port was 60 mL / min, purge time was 1 min, helium was used as carrier gas, injection port temperature was 230℃, and flow rate was 1.0 mL / min. Mass spectrometry conditions: electron impact (EI ionization mode), ion source temperature 230℃, scan mode MS2, mass scan range m / z 29~300, electron energy mode was tuned.

[0060] (2) Data processing: The data was processed using a MassHunter chemistry workstation. Identification of flavor compounds was based on retention indices, determined by comparison with retention indices of homologous reference samples (C7-C40) of normal alkanes, and by comparison with retention indices from the NIST17 and NIST20 mass spectrometry libraries. The content of volatile flavor compounds in different cheeses was calculated semi-quantitatively using information acquired by GC-MS with the addition of the internal standard compound 2-methyl-3-heptanone. Finally, 29 sets of data were obtained as the original dataset, containing 66 flavor compounds from cheeses aged 3, 6, 9, 10, and 12 months.

[0061] Data balancing: Based on 29 original datasets, synthetic minority oversampling (SMOTE) was used to synthesize partial data within the range of the original data, resulting in a cheese flavor evaluation dataset containing 104 data points for model training.

[0062] (3) Evaluation Model Construction: XGboost, KNN, RF, SVM, ANN, and DL classification models were used to train the cheese flavor evaluation dataset. The code for these models was executed in Python 3 within JupyterLab, which was launched by invoking a command-line window within the AnacondaNavigator environment. NumPy, pandas, matplotlib, seaborn, tensorflow, and imblearn libraries were installed in the environment for building and visualizing the models. These ML models were implemented using the scikit-learn library. The dataset was randomly divided into a training set (70%) and a test set (30%) and standardized. Hyperparameter grid search was used to optimize model performance. The optimal parameter combinations for each model are shown in Table 1 below.

[0063] Table 1 Optimal parameter settings for different classification models

[0064]

[0065] (4) Feature engineering: Select the model with the highest accuracy and best performance, and combine the feature selection methods such as SelectKBest, RFE, PCA, and L1 regularization to select the flavor features that make the model have the highest accuracy and can replace the 66 flavor substances to train the model.

[0066] (5) Determination of the content of characteristic flavor substances in cheeses at different maturity stages: The model was retrained using the screened flavor features, and the interaction between different characteristic flavor substances was explored using SHAP interaction values. Then, the cheese classification process was visualized using a decision tree algorithm based on the Gini coefficient to clarify the flavor characteristics of cheddar cheese at different maturity stages.

[0067] (6) The process for preparing two types of amino acid-based fast-maturing cheese is as follows:

[0068] 1) Raw milk that has passed the inspection according to GB 19301—2010 shall be pasteurized at 65℃ for 30 minutes;

[0069] 2) After sterilization, cool the milk to 30°C, add commercial direct-inoculation starter (Chr. Hansen R707) at 200 U / t, and ferment at 30°C for 1.5 hours;

[0070] 3) After fermentation is complete, add 0.2% (w / w) of CaCl2 powder and stir well;

[0071] 4) After adding CaCl2, raise the temperature to 35℃ and add 0.1% (w / w) activated rennet. The rennet is first prepared into a 2% (w / w) solution with 1% (w / w) sterile NaCl solution, and then added to the milk after being kept at 31℃ for 25 minutes. Stir for 3-5 minutes, and let it stand for 1 hour to coagulate after adding the rennet.

[0072] 5) When clear whey separates out, cut the curd with a knife. If the cut surface is flat and smooth, this is the end point of curd. At this point, use a cheese knife to cut the curd into 1cm pieces. 3 Small clumps;

[0073] 6) Then increase the temperature by 1°C every 3-5 minutes, stirring slowly at first and then faster, until it reaches 42°C and then stops;

[0074] 7) After mixing, let stand for 40 minutes, pile the curd on one side of the cheese trough, and remove the whey. Turn and squeeze the curd every 15 minutes, turning it twice to promote the removal of whey.

[0075] 8) After chopping the stacked clumps, sprinkle 1% (w / w) salt on the surface and mix well;

[0076] 9) After mixing in the salt, divide the curd into two equal portions. Sprinkle 0.04% (w / w) of leucine and tyrosine onto the surface of the broken curd pieces in one portion and mix well.

[0077] 10) Place the lumps into the cleaned and disinfected pressing mold, and press them on the press for 2-4 hours;

[0078] 11) Cut the pressed cheese into small pieces and pack them into vacuum packaging bags. Use a vacuum packaging machine to vacuum pack the cheese and let it mature at 30℃ for 7 days.

[0079] Then, the content of volatile compounds in the matured cheese was determined by GC-MS, and compared with the flavor characteristics of cheeses at different maturation stages obtained in step (5) to determine its maturation stage.

[0080] Based on the above embodiments, the accuracy, precision, recall, F1 score, and ROC curve results for each model are shown in Table 2 below. The accuracy of different models is as follows: Figure 2 As shown in Figure 3, the ROC curves for different models are as follows. Figure 3 In the figure, (a) is the ROC curve of the XGboost model, (b) is the ROC curve of the SVM model, (c) is the ROC curve of the KNN model, (d) is the ROC curve of the RF model, (e) is the ROC curve of the ANN model, and (f) is the ROC curve of the DL model.

[0081] Table 2 Precision, Recall, and F1-score of different cheese maturity classification models

[0082]

[0083]

[0084] From Table 2 and Figure 2 The results show that the RF model achieved the highest accuracy (100.00%), followed by KNN (97.00%) and SVM, ANN, and DL (all 96.88%). After cross-validation, the RF model achieved a cross-validation accuracy of 93.24 ± 5.96%. All samples had a precision, recall, and F1 score of 1.00, and the ROC curves were concentrated in the upper left corner. The AUC value for each cheese maturity stage was 1.0. These results indicate that the RF model exhibits better performance, higher accuracy, and better stability.

[0085] Further, the RF model was used in combination with Selectbest, RFE, PCA, and L1 regularized feature selection methods to screen characteristic flavor substances. When the accuracy of the model was 100%, the feature selection results are shown in Table 3.

[0086] Table 3. Feature flavor screening results for each feature screening method

[0087]

[0088]

[0089] The Recursive Feature Evidence (RFE) algorithm was used to progressively remove flavor compounds with low importance scores from all flavor compounds, which were then used as input for a new RF model. In each iteration, the RF model was reconstructed and the importance scores of the remaining flavor compounds were calculated until the stopping condition of highest accuracy and fewest flavor compounds was met. Table 3 shows that after recursive feature removal using RFE, 14 flavor compounds were finally selected, including ketones (2-heptanone, 2-nonanone, 2-undecanone, and acetoin), acids (acetic acid, butyric acid, valeric acid, hexanoic acid, octanoic acid, nonanoic acid, and decanoic acid), and lactones (γ-dodecanolactone, δ-dodecanolactone, and δ-decanoic acid). Comparison with published literature on cheese flavor research revealed that these 14 flavor compounds are consistent with the key flavor compounds identified in previous studies. The SelectKBest algorithm is a feature selection method based on statistical tests. It selects the top k features with the highest scores by calculating the statistical significance between each feature and the target variable (such as chi-square test, F-test, etc.). However, this method selects more feature flavors than the RFE algorithm, and the results do not match the key flavors identified in some literature. The PCA algorithm selects 6 principal components, but the principal components of the PCA algorithm are linear combinations of the original features, which cannot directly select the original features. Therefore, although this method can reduce dimensionality, it cannot directly output feature components. The L1 regularization algorithm selects too many features. Therefore, the RFE feature selection result is chosen.

[0090] Further, SHAP interaction values ​​were used to elucidate the contributions of the 14 selected characteristic flavor compounds at each cheese maturation stage. A cheese maturation stage evaluation model was used to conduct SHAP interaction analysis on the characteristic flavor compounds and their contents at different maturation stages, calculating SHAP interaction values ​​and generating the contribution of characteristic flavor compounds at different maturation stages, as shown in the figure. Figure 4-8 As shown, each point represents an observation in the corresponding training set. Each feature is arranged from top (left) to bottom (right) according to its importance in the current classification. The influence of a single feature is concentrated on the diagonal, while interactions occur outside the diagonal. The scatter plots are colored from blue to red, meaning feature values ​​range from low to high. Negative values ​​to the left of the central axis indicate a negative impact on the model, while those on the right indicate a positive impact. The more concentrated the scatter plots, the more consistent the impact on the model. The decision tree results for determining the cheese's maturity stage are shown in Table 4 and... Figure 9 In each decision tree, the shortest path for each cheese at its maturity stage is used to determine the flavor compounds and their content in cheeses at different maturity stages. Figure 4 The depth of the color indicates the accuracy of classifying a particular category; the darker the color, the higher the accuracy. Gini represents the Gini impurity; the smaller the Gini value, the more accurate the classification. Samples indicates the number of samples in the current node. Value indicates the possible categories of the current node and the number of samples in each category. Class represents the category with the most samples in the current node, i.e., the classification result.

[0091] Table 4. Quantitative range of potential flavors for cheeses at different maturity levels obtained from the decision tree.

[0092]

[0093]

[0094] The values ​​in Table 4 are the feature content calculated by the RF model based on the dataset, in μg / kg; the slashes indicate that they do not appear when determining cheese maturity using the shortest path; theoretically, any combination of two characteristic flavor compounds in each column can determine the maturity of that column, except for 12 months. 12-month-old cheese can be determined using any one of the three flavor compounds.

[0095] According to Table 4 and Figure 4 The presence of 2-heptanone (0–44.73 μg / kg) and δ-dodecanolide (0–5.21 μg / kg), or 2-heptanone and δ-decanolide (0–13.90 μg / kg), is commonly used to determine if a cheese is 3 months old. A hexanoic acid content of 383.45–702.75 μg / kg can directly classify it as 3 months old cheese. Figure 4 As shown, these are the characteristic flavor compounds of 3-month-old cheese.

[0096] exist Figure 5 In this study, high levels of acetic acid, δ-dodecanolactone, and 2-undecanoyl, combined with low levels of valeric acid and 2-heptanone, helped identify cheeses aged 6 months. Notably, the presence of both high levels of acetic acid and δ-decanoyl lactone enhanced the reliability of classifying cheeses as 6 months. It is also noteworthy that when the acetic acid content exceeded 26228.99 μg / kg, it surpassed the observation range for other maturities, making it the dominant characteristic flavor of 6-month-old cheeses. Acetic acid, when combined with γ-dodecanolactone (>5.83 μg / kg), δ-dodecanolactone (>7.04 μg / kg), valeric acid (0–2.49 μg / kg), and acetoin (>3.20 μg / kg), respectively, indicated 6-month maturity of cheese, as shown in Table 4. The combination of valeric acid (0–2.49 μg / kg) and 2-nonanone (>21.95 μg / kg) was also common in determining 6-month maturity of cheese. Figure 5 As shown.

[0097] For cheeses that have matured for 9 months, such as Figure 6As shown in Table 4, low levels of decanoic acid (0–498.92 μg / kg) are a prominent characteristic flavor of 9-month-old cheese. Low levels of 2-undecanoic acid (0–6.64 μg / kg) and δ-dodecanoic acid (0–10.66 μg / kg), as well as low levels of 2-undecanoic acid (0–6.64 μg / kg) and high levels of 2-heptanone (>75.86 μg / kg), also help in evaluating 9-month-old cheese, but less frequently. When 2-undecanoic acid is combined with δ-dodecanoic acid, 2-heptanone, and butyric acid (>267.68 μg / kg), it can generally be identified as 9-month-old cheese, such as... Figure 6 As shown.

[0098] Elevated levels of δ-dodelactone and δ-decalactone, and low levels of valeric acid and 2-heptanone are frequently observed in cheese produced after 10 months of processing. Figure 7 As shown in Table 4, the identification of 10-month-old cheeses is more favorable when these four flavor compounds are randomly combined with 2-undecanoic acid, butyric acid, and hexanoic acid. The most common combinations for evaluating 10-month-old cheeses are δ-dodecanolactone (>6.11 μg / kg) and δ-decanoic acid (>19.42 μg / kg), 2-undecanoic acid (≤5.43 μg / kg) and butyric acid (≤376.38 μg / kg), as shown in Table 4. Figure 7 As shown.

[0099] High levels of decanoic acid, 2-heptanone, 2-undecanone, and caprylic acid contribute positively to the identification of 12-month-old cheeses. Figure 8 As shown. In particular, when the contents of 2-heptanone, caprylic acid, and decanoic acid exceed 116.88 μg / kg, 659.02 μg / kg, and 574.55 μg / kg respectively, any one of these can be directly identified as cheese that has matured for 12 months, such as... Figure 8 As shown.

[0100] The volatile flavor results of the two types of fast-maturing cheeses are shown in Table 5.

[0101] Table 5 Volatile flavor compounds in fast-maturing cheeses

[0102]

[0103]

[0104] Based on the characteristic flavor profiles of cheeses at different maturation stages, Leu cheese may have a maturation period of 6 months because its alpha-carotene content is greater than 462.92 μg / kg and its butyric acid content is less than 15365.15 μg / kg (e.g., ...). Figure 5 (and Table 4). Tyr cheese may have a 3-month maturation period, with an hexanoic acid content of 458.18 μg / kg, ranging from 383.45 to 702.75 μg / kg (characteristics of 3-month cheese).

[0105] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0106] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating cheese maturity based on machine learning and GC-MS, characterized in that, Includes the following steps: Step 1: Collect flavor compound data of cheese samples at different maturity stages using gas chromatography-mass spectrometry and construct a flavor compound dataset; Step 2: Perform augmentation preprocessing on the flavor compound dataset to obtain the training dataset; Step 3: Build a machine learning classification model, and use the training dataset to train and optimize it to obtain a cheese maturity evaluation model; Step 4: Optimize the types of input flavor compounds for the cheese ripening period evaluation model using the feature selection method based on the training dataset, and determine the characteristic flavor compounds; Step 5: The optimized cheese ripening evaluation model outputs a decision tree, displaying the characteristic flavor compounds and their contents corresponding to cheeses at different ripening stages; Step 6: Collect characteristic flavor substance data of the cheese to be evaluated and input them into the optimized cheese maturity evaluation model to obtain the cheese maturity evaluation results.

2. The method for evaluating cheese maturity based on machine learning and GC-MS according to claim 1, characterized in that, The flavor compound dataset includes the types and corresponding contents of flavor compounds; the flavor compound dataset is processed using synthetic minority oversampling technique to obtain the training dataset.

3. The method for evaluating cheese maturity based on machine learning and GC-MS according to claim 1, characterized in that, In step 3, multiple machine learning classification models are constructed and trained separately. The machine learning classification model with the highest accuracy and best performance is selected as the cheese maturity evaluation model. The model parameters are optimized using parameter tuning methods.

4. The method for evaluating cheese maturity based on machine learning and GC-MS according to claim 1, characterized in that, Machine learning classification models include extreme gradient boosting tree models, K-nearest neighbor models, random forest models, support vector machine models, artificial neural network models, and deep learning models.

5. The method for evaluating cheese maturity based on machine learning and GC-MS according to claim 3, characterized in that, Parameter tuning methods include grid search, random search, Bayesian optimization, and automated parameter tuning.

6. The method for evaluating cheese maturity based on machine learning and GC-MS according to claim 1, characterized in that, Feature selection methods include SelectKBest, recursive feature deletion, PCA, LASSO, and L1 regularization.

7. The method for evaluating cheese maturity based on machine learning and GC-MS according to claim 1, characterized in that, In step 4, the stopping condition for model optimization is set to the number of remaining flavor compounds reaching the expected amount, or the performance index of the cheese maturity evaluation model on the test set of the training dataset no longer improving.

8. The method for evaluating cheese maturity based on machine learning and GC-MS according to claim 1, characterized in that, The cheese maturity evaluation model uses a decision tree algorithm based on information gain or Gini coefficient to generate a decision tree. The shortest path of each cheese at different maturity stages in the decision tree is used to determine the characteristic flavor compounds and their contents.

9. The method for evaluating cheese at different maturity stages based on machine learning and GC-MS according to claim 1, characterized in that, Fourteen characteristic flavor compounds were screened, including ketones, acids, and lactones. The ketones included 2-heptanone, 2-nonanone, 2-undecanone, and acetoin; the acids included acetic acid, butyric acid, valeric acid, hexanoic acid, caprylic acid, nonanoic acid, and decanoic acid; and the lactones included γ-dodecanolactone, δ-dodecanolactone, and δ-decylolactone.

Images