Liupao tea aroma type identification method and identification system

Abstract

The invention belongs to the technical field of tea quality evaluation and detection, and relates to a Liupao tea aroma type identification method and identification system. The Liupao tea aroma type identification method comprises the following steps: collecting volatile substance information in a Liupao tea sample; constructing an extreme gradient lifting model based on the volatile substance information; carrying out interpretability analysis on the extreme gradient boosting model, and screening fragrance type difference components according to the contribution degree of each feature to model prediction; and taking the screened aroma type difference components as input characteristics, retraining the extreme gradient boosting model to obtain an optimized aroma type identification model, and performing aroma type identification on the Liupao tea by adopting the optimized aroma type identification model. According to the method and the system for identifying the aroma types of the tea leaves, the quality of the Liupu tea can be simply, efficiently, objectively and accurately identified.

Description

Technical Field

[0001] This invention belongs to the field of tea quality evaluation and testing technology, specifically relating to a method and system for identifying the aroma of Liubao tea. Background Technology

[0002] Liubao tea originated in Wuzhou City, Guangxi Zhuang Autonomous Region. It is a distinctive dark tea of ​​Guangxi, made from Guangxi large-leaf variety, medium-leaf variety, and local tea tree cultivars from Cangwu County. Camellia sinensis Made from fresh leaves of (L.) O. Kuntze. Drinking Liubao tea offers multiple health benefits, including the antioxidant, blood sugar lowering, anti-hyperlipidemia, and anti-obesity effects common to black tea. It has also been shown to improve metabolic syndrome and exert anti-inflammatory effects, and is currently widely sold in China and Southeast Asia.

[0003] Aroma is a core indicator for evaluating the quality and market value of Liubao tea. Liubao tea boasts a rich variety of aroma types, a characteristic closely related to microbial fermentation during processing (such as the metabolic activity of microorganisms like Aspergillus and yeast) and the transformation of active ingredients and accumulation of flavor substances during subsequent aging. Currently, the two main aroma types in commercially available Liubao tea are aged aroma and areca nut aroma. Aged aroma is based on aged, woody, and caramel notes, complemented by fruity and floral aromas, resulting in a mellow and rich fragrance with a sweet taste. Areca nut aroma is characterized by a unique minty aroma, combined with aged, floral, fruity, smoky, and spicy notes, creating a rich and layered flavor. It is worth noting that areca nut aroma requires long-term aging to develop, but its intensity fluctuates greatly depending on aging time and tea grade, making it incomparable to the aroma of areca nut fruit. Currently, the industry's descriptions of these two aroma types remain vague, and professional evaluation resources are scarce, posing a significant challenge to distinguishing between them.

[0004] The aroma characteristics of tea are directly related to its chemical composition, with volatile organic compounds (VOCs) being the core material basis. Studies have shown that the VOCs in Liubao tea mainly include hydrocarbons, aldehydes, alcohols, and ketones: key aroma compounds in aged tea include 1-methylnaphthalene, decanal, β-ionone, nonanal, cedrol, β-linalool, dihydroactinol, and α-terpineol; important aroma compounds in areca nut tea include 14 compounds such as cedrol, α-terpineol, linalool and its oxides, phenethyl alcohol, and dehydro-β-ionone. Existing research indicates that aroma formation is synergistically driven by multiple VOCs, and some key substances overlap between aroma types, suggesting potential chemical connections and transformation between aroma types. This complexity further increases the difficulty of accurately identifying different aroma types of Liubao tea through traditional sensory evaluation. Summary of the Invention

[0005] Based on the above technical problems, the present invention provides a method for identifying the aroma type of Liubao tea, which can provide technical support for accurately identifying the aroma type of Liubao tea.

[0006] This invention is achieved through the following technical solution: In a first aspect, this invention provides a method for identifying the aroma type of Liubao tea, comprising the following steps: Information on volatile substances in Liubao tea samples was collected. An extreme gradient boosting model is constructed based on the information on the volatile substances. An interpretability analysis was performed on the extreme gradient boosting model, and the aroma-differentiating components were screened based on the contribution of each feature to the model prediction. Using the selected aroma difference components as input features, the extreme gradient boosting model is retrained to obtain an optimized aroma identification model. The optimized aroma identification model is then used to identify the aroma of Liubao tea.

[0007] As a preferred embodiment of the present invention, the optimization process of the extreme gradient boosting model includes the following steps: The hyperparameters of the extreme gradient boosting model are optimized using a grid search and cross-validation method. An extreme gradient boosting tree model is then established based on the optimized parameters. Interpretability analysis is performed on the extreme gradient boosting tree model to screen for differential components that make significant contributions to aroma classification. Based on these differential components, an optimized extreme gradient boosting model is obtained by remodeling.

[0008] More preferably, the optimized parameters are: a learning rate of 0.05, a maximum tree depth of 3, a minimum loss reduction required for node splitting of 0, a column sampling ratio of 0.8 per tree, a minimum sum of sample weights in child nodes of 1, and an L2 regularization term of 1.

[0009] In a preferred embodiment of the present invention, the volatile substance information is acquired using gas chromatography-ion mobility spectrometry and an electronic nose.

[0010] More preferably, the parameters for electronic nose acquisition are set as follows: preheating for 30 min, gas washing until the sensor response value stabilizes at 1.0±0.05, gas flow rate of 300 mL / min, detection time of 70.0 s, gas washing time of 70 s, and data from 63 to 66 s are selected.

[0011] More preferably, the aroma-differentiating components screened based on information collected by the electronic nose are sulfides, terpenes, nitrogen oxides, and alkanes.

[0012] More preferably, the parameters for gas chromatography-ion mobility spectrometry acquisition are set as follows: GC conditions: injection volume 500 μL, injection port temperature 80℃, injection needle temperature 85℃, column temperature 60℃ isothermal; carrier gas program settings are as follows: initial flow rate 2.0 mL / min held for 2 min, flow rate linearly increased to 10.0 mL / min within 8 min, flow rate linearly increased to 100.0 mL / min within 10 min, held for 40 min; IMS conditions: positive ion mode, ³H ionization source, migration tube temperature 45℃, drift gas flow rate 75.0 mL / min.

[0013] More preferably, the aroma-differentiating components screened using information collected by gas chromatography-ion mobility spectrometry are diallyl sulfide-D, acetal, 2-methylpyridine, and α-pinene.

[0014] The present invention also provides a Liubao tea aroma identification system, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the Liubao tea aroma identification method as described in any of the above technical solutions.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a method for identifying the aroma type of Liubao tea, enabling simple, objective, and accurate identification and differentiation of its aroma types. The method eliminates the need for complex sample pretreatment. It acquires aroma information from the tea liquor and dry tea using an electronic nose or from the dry tea using gas chromatography-ion mobility spectrometry. Using this information as input, a grid search combined with cross-validation is employed to optimize the hyperparameters of an extreme gradient boosting model. An extreme gradient boosting tree model is then established based on the optimized parameters. Interpretability analysis of the optimized extreme gradient boosting tree model identifies key differential components that significantly contribute to aroma type classification. Finally, using these key aroma differential components as input variables, an extreme gradient boosting model for Liubao tea aroma identification is established, reducing the impact of redundant features on the model and improving its prediction efficiency. Attached Figure Description

[0016] Figure 1 This is a radar diagram showing the response of Liubao tea leaves and tea liquor in various electronic nose channels. A, Dry Tea Group; B, Tea Liquor.

[0017] Figure 2 This is a GC-IMS differential spectrum of volatile components in the sample.

[0018] Figure 3 These are fingerprint spectra of volatile compounds in Liubao tea of ​​different aroma types from different manufacturers. A) Liubao tea sample from Wuzhou Zhongcha; B) Liubao tea sample from Guangxi Wuzhou Tea Factory.

[0019] Figure 4This is a ranking chart of feature importance based on the extreme gradient boosting model of interpretable analysis values; A, dry tea group, B, tea soup group.

[0020] Figure 5 The results are based on the extreme gradient boosting model using GC-IMS data; (A) training set confusion matrix; (B) test set confusion matrix.

[0021] Figure 6 (A) is a ranking of the importance of features in the extreme gradient boosting model based on interpretability analysis; (B) is a global interpretation graph of the interpretability analysis values. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] 1. Materials and Methods 1.1 Materials and Reagents Nine aged aroma types and nine areca nut aroma types of Liubao tea samples were purchased from two different manufacturers in Guangxi: Wuzhou Zhongcha Tea Industry Co., Ltd. and Guangxi Wuzhou Tea Factory Co., Ltd. The Liubao tea samples included all grades of tea from both aroma types produced by different tea factories. After collection, the tea samples were stored at -20℃ in a sealed, light-protected environment. Before the experiment, they were brought to room temperature (25±1℃) to ensure the stability of the aroma components.

[0024] The n-ketones used in this experiment (2-butanone, 2-pentanone, 2-hexanone, 2-heptanone, 2-octanone, and 2-nonanone) were all analytically pure and sourced from Aladdin Company, used to establish retention time-retention index calibration curves. High-purity nitrogen (≥99.999%) was used as the GC carrier gas and IMS drift gas to avoid interference from impurity gases in the detection of volatile components. Distilled water was sourced from Watsons Guangzhou Co., Ltd., used for tea brewing experiments to eliminate the influence of water quality on aroma components.

[0025] 1.2 Experimental Apparatus Electronic nose system: PEN 3.5 electronic nose (equipped with 10 selective metal-oxide-semiconductor sensors, WinMuster Airsense Analytics Inc., Germany). Details of the 10 selective metal-oxide-semiconductor sensors are shown in Table 1. Auxiliary equipment: 20 mL brown headspace vials, GB / T 23776-2018 standard cylindrical evaluation cups, PTFE sealing film, constant temperature water bath (temperature control ±0.5℃), electronic balance (accuracy 0.01 g, Shanghai Puchun Metrology Instrument Co., Ltd.).

[0026] Table 1. Detailed information on 10 selective metal-oxide-semiconductor sensors for the electronic nose GC-IMS system: CTC-PAL 3 static headspace autosampler (CTC Analytics AG, Switzerland), FlavorSpec® gas ion mobility spectrometer (with ³H ionization source, GAS, Germany), MXT-WAX capillary column (30 m × 0.53 mm × 1.0 μm, Restek, USA), analytical balance (accuracy 0.0001 g, Sartorius Scientific Instruments Ltd.).

[0027] 2. Experimental Methods 2.1 Electronic nose analysis Dry tea processing method: Weigh 1.00 g of tea sample into a 20 mL brown headspace bottle, seal it, heat it in a water bath at 65℃ for 20 min, and cool it to room temperature before testing.

[0028] The tea samples were treated using the brewing method: according to GB / T 23776-2018, 1.00 g of tea sample was weighed into a 150 mL evaluation cup, brewed with 50 mL of 100℃ distilled water for 2 min, and the tea liquor was discarded. After a second brewing for 5 min, the tea was filtered out, and 5 mL of the tea liquor was placed in a headspace bottle, sealed, and cooled for testing.

[0029] Preheat for 30 min (clean air environment: 25±1℃, humidity 50±5%), and wash the gas until the sensor response value stabilizes at 1.0±0.05. Gas flow rate is 300 mL / min, detection time is 70.0 s, and gas washing time is 70 s; data from 63 to 66 s are selected, and each sample is measured in triplicate.

[0030] 2.2 GC-IMS Analysis Weigh 1.000 g of tea sample into a 20 mL brown headspace vial, seal it, and incubate at 80 °C for 20 min (500 r / min). Inject the sample immediately.

[0031] GC conditions: splitless injection, injection volume 500 μL, injection port temperature 80℃, injection needle temperature 85℃, column temperature isothermal 60℃. Carrier gas (high-purity nitrogen) program settings are as follows: initial flow rate 2.0 mL / min, held for 2 min, linearly increase the flow rate to 10.0 mL / min within 8 min, linearly increase the flow rate to 100.0 mL / min within 10 min, and hold for 40 min.

[0032] IMS conditions: positive ion mode, ³H ionization source, migration tube temperature 45℃, drift gas flow rate 75.0 mL / min (high-purity nitrogen). Each sample was measured in triplicate.

[0033] 2.3 Data Analysis In the electronic nose analysis, the data was opened in Excel, and data from 63 to 66 seconds were selected for subsequent analysis. GC-IMS data were analyzed and matched for substances using VOCal 0.4.03 software from GAS GmbH, Dortmund, Germany, and information on volatile components was exported.

[0034] The electronic nose and GC-IMS data were analyzed separately. The imported data format was .xlsx, which included sample number information, sample category label information, types of volatile substances, and response information. The data analysis process was performed in Python. First, the necessary libraries for data analysis were imported, including pandas, nump, matplotlib, seaborn, sklearn, and xgboost. The read_excel function was used to read the data. The data was then divided into training and test sets with a ratio of 70% for the training set and 30% for the test set. Subsequently, data standardization preprocessing was performed on both the training and test set data.

[0035] Based on the training set data, a grid search combined with cross-validation is used to optimize the hyperparameters of the extreme gradient boosting model, obtaining the optimal modeling parameters and saving the best model. The model can then be loaded and applied, and its performance is evaluated and visualized using training and test set data. Evaluation metrics include accuracy, precision, recall, F1 score, ROC curve, and Kappa coefficient. Finally, interpretability analysis is performed on the model using a tree interpreter to calculate the interpretable values ​​of each variable, quantifying the contribution of each feature to the model's classification and thus identifying important classification features.

[0036] 3. Experimental Results 3.1 Information on volatile components obtained from electronic nose analysis Figure 1 Electronic nose response radar map, distributed as dry tea (A) and tea infusion (B), from Figure 1 As can be seen, the differences in response values ​​between the aroma types of Liubao tea under the two treatment methods are mainly reflected in three sensors: the W5S sensor (sensitive to nitrogen oxides), the W1W sensor (sensitive to organic sulfur and terpene aromatics), and the W2W sensor (sensitive to organic sulfur compounds and terpene aromatics). Among them, the response value of the W1W sensor channel is the highest, followed by the W2W and W5S sensors. In addition, it was found that the response values ​​of areca nut aroma Liubao tea on the above sensors are generally higher than those of aged aroma Liubao tea. This indicates that the concentration of volatile compounds in areca nut aroma Liubao tea is higher than that in aged aroma Liubao tea, and also confirms the feasibility of distinguishing Liubao tea aroma types based on electronic nose response.

[0037] 3.2 Information on volatile components obtained from GC-IMS analysis like Figure 2 As shown, most signal peaks in the GC-IMS analysis of Liubao tea samples appeared in the retention time range of 200–800 s and the drift time range of 1.0–1.5 s. Furthermore, the contour lines of CG (Guangxi Wuzhou Tea Factory aged Liubao tea sample), CZ (Guangxi Wuzhou Tea Factory Zhongcha aged Liubao tea sample), BG (Guangxi Wuzhou Tea Factory areca nut aroma Liubao tea sample), and BZ (Wuzhou Zhongcha areca nut aroma Liubao tea sample) in the topographic map showed a generally similar distribution, indicating that the two aroma types of Liubao tea have similar compound types. Similar to the phenomenon observed in electronic nose analysis, GC-IMS analysis also showed that the volatile compound response level of the areca nut aroma tea sample was higher than that of the aged aroma tea sample. In addition, the intensity and number of peaks representing volatile organic compounds in samples CZ and BZ were greater than those in CG and BG, indicating that Liubao tea from manufacturer Z contained a higher variety and intensity of volatile organic compounds. The differences between the sample groups provide a basis for further differentiation of the aroma types of Liubao tea.

[0038] By searching and comparing the GC Retention Index Database (NIST) and IMS Migration Time Database built into the VOCal software, a total of 148 signal peaks were detected in the Liubao tea samples, and 131 peaks, comprising 117 compounds, were identified. Detailed information on these compounds is listed in Table 1. Based on their chemical structures, they can be divided into 10 categories: 28 aldehydes (23.93%), 22 ketones (18.80%), 17 alcohols (14.53%), 15 organic heterocyclic compounds (12.82%), 11 terpenes (9.40%), 10 esters (8.55%), 8 organosulfur compounds (6.84%), 3 hydrocarbons (2.56%), 2 acids (1.71%), and 1 inorganic compound (ammonia, 0.85%). Overall, aldehydes, ketones, and alcohols are substances with relatively high volatile content in aged and areca nut-flavored Liubao teas.

[0039] Figure 3 This is a fingerprint spectrum of volatile compounds in aged and areca nut-flavored Liubao teas, with each point representing the peak intensity of a specific compound in the corresponding sample. The figure presents the composition of volatile organic compounds in each sample and the differences in volatile organic compound content between samples. As shown in Figures 3A and 3B, 3-furanmethanol, furfural, (E)-2-hexenal-M, o-xylene, 1-penten-3-ol-M, diallyl sulfide-M, hexanal-M, 1-penten-3-one-M, 2-butanone, acetal, butanal, acetone, and propanal were detected in Liubao tea at relatively high levels. These substances are mainly aldehydes, ketones, and alcohols, and can be considered characteristic volatile compounds of post-fermented dark tea. In addition, acetone, acetic acid, 2-butanone, 2-propanol, hexanal, 6-methyl-5-hepten-2-one, and (E)-2-hexenal were also identified. These substances have been reported to be present in high concentrations in dark tea. Specific volatile components are shown in Table 2.

[0040] Table 2. List of volatile components in the sample Note: "D" indicates a dimer; "M" indicates a monomer; "MW" indicates a monomer. a "Molecular weight"; "RI" b "Retention index in capillary gas chromatography column; "Rt c "Retention time in capillary gas chromatography column; "Dt d "Drift time in the drift tube (relative to RIP)".

[0041] from Figure 3As can be seen, the content of compounds differs among different aroma types of Liubao tea. These differences are marked with red boxes, and significance analysis was performed to assess their potential as aroma differentiation markers. Independent samples t-tests of these compounds showed that 21 compounds in the Guangxi Wuzhou Tea Factory (G) sample had significant differences in content among different aroma types of Liubao tea, while 37 substances in the Wuzhou Zhongcha (Z) sample showed significant differences within the group (p<0.05). Among them, (Z)-4-decenal, decanal, cis-2-penten-1-ol, 2-heptanone, diallyl sulfide-D, propylacetate, 2-propenal, and 2-methylpyridine were aroma differentiation compounds common to samples from both manufacturers. It is worth noting that 2-heptanone, a ketone compound with pear-like and slightly medicinal aroma, is present in higher amounts in areca nut-scented Liubao tea; diallyl sulfide-D, an organosulfur compound, is present in higher amounts in aged tea.

[0042] 3.3 Key Feature Screening and Aroma Identification Model Construction Based on Extreme Gradient Boosting Method 3.3.1 Electronic nose data analysis After importing the electronic nose data, the training and test sets were divided in a 7:3 ratio. A grid search and cross-validation method was used to optimize the hyperparameters of the extreme gradient boosting model. The optimized parameters were: learning rate of 0.05, maximum tree depth of 3, gamma=0, colsample_bytree=0.8, min_child_weight=1, and reg_lambda=1. Based on these optimal parameters, an extreme gradient boosting tree model was built, saved, and evaluated on both the training and test sets. Accuracy, precision, and recall were used to evaluate the model's performance. The evaluation results on the training and test sets are shown in Table 3. Based on the evaluation parameters, the model constructed from the electronic nose data of the tea infusion group showed better identification of the aroma type of Liubao tea.

[0043] Table 3 Evaluation results of the extreme gradient boosting model based on the electronic nose sensor channel on the training and test sets. Note: In the table, "Model-Raw train" refers to the training model built based on all variables, and "Model-Raw test" refers to the prediction of the training model based on all variables on the test set samples. "Model-F1 train" refers to the training model built based on the variables ranked first in importance according to interpretability analysis, and "Model-F1 test" refers to the prediction of the Model-F1 train model on the prediction set samples. Similarly, "Model-F2 train" refers to the training model built based on the variables ranked first and second in importance according to interpretability analysis, and "Model-F2 test" refers to the prediction of the Model-F2 train model on the prediction set samples.

[0044] Subsequently, an interpretability analysis was conducted on the extreme gradient boosting model to identify key feature variables that significantly contribute to aroma classification, based on the feature importance plot ( Figure 4 As shown in Table 3, W1C contributes the most to the model classification in the dry tea group, while W1W and W5S make significant contributions to the model classification in the tea infusion group, exhibiting higher mean absolute interpretability values. These important feature variables were selected for remodeling to verify their importance. A stepwise variable addition method was adopted until the recognition accuracy no longer increased. The results are shown in Table 3. When the dry tea group data was remodeled using the first two sensor channels W1C (aromatic compounds) and W2W (aromatic components and organosulfur compounds), the resulting model was better than the original modeling result using all dry tea features. When the tea infusion data was remodeled using the first three sensor channels W1W (sulfides and terpenes), W5S (nitrogen oxides), and W3S (alkanes), the resulting model was better than the original modeling result using all tea infusion features. These results indicate that these variables are key features for aroma identification.

[0045] Based on the selected feature variables, an electronic nose-based aroma identification model for Liubao tea was established by integrating important feature variables of dry tea and tea infusion (dry tea W1C, W2W; tea infusion W1W, W5S, W3S). The results are shown in Table 3. The results of the remodeling based on the selected variables are the same as the results of the original variables in the tea infusion group, indicating that the aroma components of tea infusion selected in the electronic nose analysis are more effective than those of dry tea in aroma identification. The identification model established based on the feature variables selected in the tea infusion group can be selected for aroma identification of Liubao tea. The aroma identification accuracy rate for the test set samples is 82.4%.

[0046] 3.3.2 GC-IMS Data Analysis After importing the GC-IMS data, the training and test sets were divided in a 7:3 ratio. A grid search and cross-validation method was used to optimize the hyperparameters of the extreme gradient boosting model. The optimized parameters were: learning rate of 0.05, maximum tree depth of 3, gamma=0, colsample_bytree=0.8, min_child_weight=1, and reg_lambda=1. Based on these optimal parameters, an extreme gradient boosting model was built, saved, and evaluated on both the training and test sets. Accuracy, precision, and recall were used to evaluate the model's performance. The evaluation results on the training and test sets are shown in Table 4. Compared to the analysis results of the electronic nose data, the GC-IMS data provides richer data information, and the confusion matrix of the established model (…) Figure 5 Figures A and B show that the samples in both the training and test sets can be accurately classified, and the model's accuracy and precision parameters are all 1, demonstrating excellent aroma classification ability.

[0047] Further interpretability analysis was conducted on the extreme gradient boosting model to identify key feature variables that significantly contribute to aroma classification, based on the feature importance plot ( Figure 6 As shown in A), diallyl sulfide-D has the most significant contribution to the model classification, exhibiting the largest mean absolute interpretability value. Acetal is the second most important classification contributing feature. Figure 6 The global interpretation plot of the B interpretability analysis also shows that diallyl sulfide-D and acetal contribute in opposite directions to the classification of the two aroma types of Liubao tea. Further, a stepwise variable addition method was adopted, selecting these important feature variables for remodeling to verify their importance until the recognition accuracy no longer increased. Finally, when four feature variables—diallyl sulfide-D, acetal, 2-methylpyridine, and alpha-pinene—were selected for remodeling, the performance of the resulting model was the same as the original model using full feature modeling, with a 100% accuracy rate in aroma type recognition, demonstrating the effectiveness of these four features.

[0048] Table 4. Model evaluation results based on GC-IMS data Note: In the table, "Model-Raw train" refers to the training model built based on all variables, and "Model-Raw test" refers to the prediction of the training model based on all variables on the test set. "Model-F2 train" refers to the training model built based on the two most important variables selected through interpretability analysis, and "Model-F2 test" refers to the prediction of the Model-F2 train model on the prediction set. Similarly, "Model-F3 train" refers to the training model built based on the three most important variables selected through interpretability analysis, and "Model-F3 test" refers to the prediction of the Model-F3 train model on the prediction set. "Model-F4 train" refers to the training model built based on the four most important variables selected through interpretability analysis, and "Model-F4 test" refers to the prediction of the Model-F4 train model on the prediction set.

[0049] This invention utilizes an electronic nose and GC-IMS data analysis, combined with an extreme gradient boosting model and interpretability analysis, to identify key variables for the identification of Liubao tea aroma types. A model built using these key variables can achieve accurate identification of Liubao tea aroma types. Specifically, the model based on key characteristic variables from the electronic nose's tea aroma information achieved an accuracy rate of 82.4%. Furthermore, GC-IMS analysis identified four key volatile components: diallyl sulfide-D, acetal, 2-methylpyridine, and α-pinene, enabling accurate identification of Liubao tea aroma types with a 100% accuracy rate.

[0050] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and scope of protection of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the content of this specification should be included within the scope of protection of the present invention.

Claims

1. A method for identifying the aroma type of Liubao tea, characterized in that, Includes the following steps: Information on volatile substances in Liubao tea samples was collected. An extreme gradient boosting model is constructed based on the information on the volatile substances. An interpretability analysis was performed on the extreme gradient boosting model, and the aroma-differentiating components were screened based on the contribution of each feature to the model prediction. Using the selected aroma difference components as input features, the extreme gradient boosting model is retrained to obtain an optimized aroma identification model. The optimized aroma identification model is then used to identify the aroma of Liubao tea.

2. The method for identifying the aroma type of Liubao tea according to claim 1, characterized in that, The optimization process of the extreme gradient boosting model includes the following steps: The hyperparameters of the extreme gradient boosting model are optimized using a grid search and cross-validation method. An extreme gradient boosting tree model is then established based on the optimized parameters. Interpretability analysis is performed on the extreme gradient boosting tree model to screen for differential components that make significant contributions to aroma classification. Based on these differential components, an optimized extreme gradient boosting model is obtained by remodeling.

3. The method for identifying the aroma type of Liubao tea according to claim 2, characterized in that, The optimized parameters are: learning rate of 0.05, maximum tree depth of 3, minimum loss reduction required for node splitting of 0, column sampling ratio of 0.8 per tree, minimum sum of sample weights in child nodes of 1, and L2 regularization term of 1.

4. The method for identifying the aroma type of Liubao tea according to claim 1, characterized in that, The volatile substance information was acquired using gas chromatography-ion mobility spectrometry or an electronic nose.

5. The method for identifying the aroma type of Liubao tea according to claim 4, characterized in that, The parameters for electronic nose data acquisition were set as follows: preheating for 30 min, gas washing until the sensor response value stabilized at 1.0±0.05, gas flow rate of 300 mL / min, detection time of 70.0 s, gas washing time of 70 s, and data from 63 to 66 s were selected.

6. The method for identifying the aroma type of Liubao tea according to claim 5, characterized in that, The aroma-differentiating components identified by information collected from the electronic nose are sulfides, terpenes, nitrogen oxides, and alkanes.

7. The method for identifying the aroma type of Liubao tea according to claim 4, characterized in that, The parameters for gas chromatography-ion mobility spectrometry acquisition were set as follows: GC conditions: injection volume 500 μL, injection port temperature 80℃, injection needle temperature 85℃, column temperature 60℃ isothermal; carrier gas program settings are as follows: initial flow rate 2.0 mL / min held for 2 min, flow rate linearly increased to 10.0 mL / min within 8 min, flow rate linearly increased to 100.0 mL / min within 10 min, held for 40 min; IMS conditions: positive ion mode, ³H ionization source, migration tube temperature 45℃, drift gas flow rate 75.0 mL / min.

8. The method for identifying the aroma type of Liubao tea according to claim 7, characterized in that, The aroma-differentiating components identified using information obtained from gas chromatography-ion mobility spectrometry were diallyl sulfide-D, acetal, 2-methylpyridine, and α-pinene.

9. A system for identifying the aroma type of Liubao tea, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the method for identifying the aroma type of Liubao tea as described in any one of claims 1 to 8.

Images