Chemical characteristic analysis method for sensory indexes of stimulated parts of tobacco leaves

Abstract

The invention relates to the technical field of chemical characteristic analysis and quality improvement of tobacco leaves, in particular to a chemical characteristic analysis method for sensory indexes of stimulated parts of tobacco leaves. The method comprises the following steps: acquiring tobacco leaf samples with different stimulation sensory characteristics, and acquiring information to obtain a modeling data set; taking the chemical component data, the pH value, the tobacco leaf part information and the production place information as modeling indexes, taking the stimulation part sensory index as a prediction target, and constructing a tobacco leaf stimulation part sensory index prediction model based on a random forest model; and based on the prediction model, analyzing the key variable in the modeling index, and determining the action direction of the key variable on the prediction target. According to the method, stimulation subdivision indexes in the sense organ are concerned as the key influencing the sense organ, great significance is achieved for exploring the material basis influencing the indexes, the limitation of general analysis of overall stimulation is broken through, and a precise target spot is provided for follow-up part specific quality regulation and control.

Description

Technical Field

[0001] This invention relates to the field of chemical characteristic analysis and quality improvement of tobacco leaves, and in particular to a method for chemical characteristic analysis of sensory indicators of irritant parts of tobacco leaves. Background Technology

[0002] Sensory indicators of tobacco leaves are an important component of the tobacco leaf evaluation system, with indicators such as nasal and oral irritation being key criteria for distinguishing the sensory quality of tobacco leaves. Meanwhile, the chemical composition of tobacco leaves (including conventional chemical components, anions and cations, and tobacco polyphenols) is a crucial material basis for determining their quality. Exploring the relationship between these two aspects helps to deepen the understanding of the formation mechanism of tobacco leaf quality, providing targets for subsequent enhancement of tobacco leaf style and quality, and promoting high-quality development in the tobacco industry.

[0003] Current research on the relationship between chemical components and sensory indicators of tobacco leaves has yielded many results: the chlorine content in Sichuan flue-cured tobacco affects volatile aroma substances and sensory quality, especially showing significant differences between different parts, and exhibits a quadratic curve relationship with most sensory indicators (Xu Zicheng, Li Dandan, Bi Qingwen, et al. Study on the relationship between chlorine content and volatile aroma substances and sensory quality of flue-cured tobacco [J]. Journal of Tobacco Science, 2008, 14(5):27-32.); increasing the content of reducing sugar, chloride ions and potassium-chlorine ratio, and decreasing the sugar-alkali ratio and nicotine content can have a positive impact on the aroma and quality of flue-cured tobacco leaves from Henan, Hunan, Zimbabwe and Brazil (Yu Jianjun, Pang Tianhe, Liu Guoshun, et al. Correlation and path analysis of aroma and quality of flue-cured tobacco with chemical components [J]. Chinese Agricultural Science Bulletin, 2006, 22(1):71-73.). Patent CN111802692A provides a method for improving the quality of tobacco leaves. The key lies in introducing magnetic Fe3O4 / porous graphene oxide-based composite aroma-enhancing particles during the preparation of the enzymatic extract. These particles contain modified chitosan loaded into their pores, and lysine microcapsules are uniformly dispersed within the chitosan, thereby more effectively improving the sensory smoking quality of the tobacco leaves.

[0004] However, the following problems still exist: (1) When using traditional statistical analysis methods to mine chemical characteristics, statistical methods such as maximum and minimum values, mean, t-test, and variance test are often used. It is difficult to analyze the positive and negative effects and importance ranking of components in detail. The identification of chemical components that affect quality is not accurate enough and lacks specificity; (2) At present, the model construction of sensory quality focuses more on indicators such as aroma and temperament, and relatively less on stimuli indicators. There is even less research on the stimulation sites; (3) In terms of the scope of chemical components, previous studies have been limited to a few compounds, which has a limited coverage and may affect the estimation accuracy of the model. Summary of the Invention

[0005] The purpose of this invention is to provide a method for analyzing the chemical characteristics of sensory indicators of irritant parts of tobacco leaves, so as to solve the above-mentioned technical problems.

[0006] This invention provides a method for analyzing the chemical characteristics of sensory indicators at irritant sites of tobacco leaves, comprising the following steps: S1. Obtain tobacco leaf samples with different sensory stimulating characteristics, collect the macro and semi-micro chemical composition data, pH value, tobacco leaf part information and origin information of each tobacco leaf sample, and after standardizing each chemical composition data, obtain the modeling dataset, and divide the modeling dataset into training set, validation set and test set according to the proportion. S2. Using the chemical composition data, pH value, tobacco leaf part information and origin information as modeling indicators, and the sensory indicators of the irritating parts as prediction targets, a mapping relationship between the modeling indicators and the prediction indicators is constructed based on a random forest model. The prediction model is trained and optimized to obtain a prediction model for the sensory indicators of the irritating parts of tobacco leaves. The sensory indicators of the irritating parts include nasal irritation indicators, oral irritation indicators and laryngeal irritation indicators. S3. Based on the optimized sensory index prediction model of the irritation site of tobacco leaf, the sensory index of the irritation site of the tobacco leaf sample is predicted, the key variables in the modeling index are analyzed, and the direction of the effect of the key variables on the prediction target is determined.

[0007] Preferably, near-infrared spectroscopy is used when collecting the macro- and semi-micro chemical composition data of each tobacco leaf sample in step S1.

[0008] Preferably, when the near-infrared prediction method acquires chemical composition data, it adopts a prediction method based on spectral conversion: acquiring the true moisture content and first near-infrared spectrum of the tobacco leaf sample, obtaining the correction parameters corresponding to the true moisture content from the pre-acquired correspondence between different moisture contents and correction parameters; converting the first near-infrared spectrum into a second near-infrared spectrum under standard moisture content according to the correction parameters; inputting the second near-infrared spectrum into the near-infrared prediction model, and outputting the chemical composition data in the tobacco leaf sample.

[0009] Preferably, training and optimizing the prediction model in step S2 includes: The method of finding the optimal discrimination threshold is adopted to improve the model prediction accuracy. The criterion for determining the optimal threshold is the F1 score of the model on the validation set. The threshold with the highest F1 score is selected as the optimal discrimination threshold to improve the problem of decreased model prediction accuracy caused by the imbalance of sensory data distribution.

[0010] Preferably, the training and optimization of the prediction model further includes: The parameters of the multinomial random forest model in the prediction model were tuned using a lattice search method. The tuning criterion was the F1 score of the model on the validation set. The parameter combination with the highest F1 score was selected as the optimal parameter. After optimization, a prediction model for sensory indicators of tobacco leaf stimulation sites was obtained. The multinomial random forest model parameters include the number of decision trees n_estimators, the maximum depth of the decision trees max_depth, the maximum number of features to split max_features, the minimum number of samples required to split internal nodes min_samples_split, and the minimum number of samples contained in a leaf node min_samples_leaf.

[0011] Preferably, the key variables in the modeling indicators analyzed in step S3 include: By ranking the variables in the random forest model by the amount of variance reduction, each feature f is calculated, which is the mean of the variance reduction of chemical composition data when splitting across all decision tree nodes. The mean result is averaged over all trees to obtain the importance of each feature f. Combined with five-fold cross-validation, the top 10 features f with the highest importance in the five calculations and appearing at least 4 times are identified as key variables related to the prediction index.

[0012] Preferably, determining the direction of the influence of the key variable on the prediction target in step S3 includes: Calculate the mean of the key variable for data groups containing sensory indicators of the stimulation site and data groups without sensory indicators of the stimulation site, and perform a mean difference test. If the mean of the key variable in the data group containing sensory indicators of the stimulation site is significantly higher than that in the data group without sensory indicators of the stimulation site, then the variable has a positive effect on the sensory indicator of the stimulation site; if the mean is significantly lower than that in the data group without sensory indicators of the stimulation site, then it has a negative effect.

[0013] Preferably, the method further includes step S4: regulating the key variables in the tobacco leaves according to the direction of action in step S3 to improve the quality of the tobacco leaves.

[0014] The present invention also provides an electronic device, comprising: Processor; and, A memory configured to store computer-executable instructions, which, when executed, cause the processor to implement the steps of the method for analyzing the chemical characteristics of sensory indicators of irritant sites in tobacco leaves as described in any one of claims 1 to 8.

[0015] The present invention also provides a storage medium for storing computer-executable instructions, which, when executed, implement the steps of the chemical characteristic analysis method for sensory indicators of tobacco leaf irritation sites as described in any one of claims 1 to 8.

[0016] Beneficial effects: This invention focuses on the detailed indicators of stimulation in the senses, namely the modeling of stimulation-sensory sites. As a key factor affecting the senses, exploring the material basis of these indicators is of great significance. It breaks through the limitations of general analysis of the overall stimulation and provides precise targets for subsequent site-specific quality control. This invention utilizes near-infrared spectroscopy to acquire chemical data, ensuring the accuracy of the acquired data, and explores the relationship between 70 chemical indicators and stimulation sites, covering a wider range of indicators. Based on model construction, this invention, combined with model analysis and hypothesis testing, further explores the key components and their effects on different stimulating sites, proposes quality improvement ideas, provides guidance for production practice, and can provide a feasible operational approach for targeted quality improvement in tobacco production. It has direct practical significance for improving the stability of tobacco sensory quality and reducing adverse stimuli. Attached Figure Description

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

[0018] Figure 1 A flowchart illustrating a method for analyzing the chemical characteristics of sensory indicators of irritant parts of tobacco leaves, provided in one or more embodiments of this specification; Figure 2 This is a schematic diagram of the structure of an electronic device provided for one or more embodiments of this specification. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0020] Method Implementation Examples According to embodiments of the present invention, a method for analyzing the chemical characteristics of sensory indicators of irritant parts of tobacco leaves is provided. Figure 1 This is a flowchart of the chemical feature analysis method according to an embodiment of the present invention.

[0021] like Figure 1 As shown, the method for analyzing the chemical characteristics of sensory indicators of irritant parts of tobacco leaves according to an embodiment of the present invention specifically includes the following steps: S1. Obtain tobacco leaf samples with different sensory stimulating characteristics, collect the macro and semi-micro chemical composition data, pH value, tobacco leaf part information and origin information for each tobacco leaf sample, and after standardizing each chemical composition data, obtain the modeling dataset, and divide the modeling dataset into training set, validation set and test set according to the proportion. The tobacco leaf samples contained 69 macro- and semi-micro chemical components, namely: water-soluble total sugar, reducing sugar, total alkaloids, total nitrogen, potassium, chlorine, starch, sulfate, phosphate, calcium, magnesium, neochlorogenic acid, chlorogenic acid, cryptochlorogenic acid, hyoscyamine, rutin, oxalic acid, malonic acid, succinic acid, malic acid, citric acid, vanillic acid, myristic acid, hexadecanoic acid, linoleic acid + oleic acid, linolenic acid, stearic acid, eicosanoic acid, aspartic acid, threonine, serine, asparagine, glutamic acid, glutamine, glycine, alanine, valine, and cysteine. Amino acids, methionine, isoleucine, leucine, tyrosine, phenylalanine, 4-aminobutyric acid, lysine, histidine, tryptophan, arginine, proline, 1-deoxy-1-L-alanine-D-fructose (FRU-ALA), 1-deoxy-1-L-valine-D-fructose (FRU-VAL), 1-deoxy-1-L-proline-D-fructose (FRU-PRO), 1-deoxy-1-L-phenylalanine-D-fructose (FRU-PHE), 1-deoxy-1-L-tryptophan 1-Deoxy-1-L-Isoleucine-D-Fructose (FRU-TRP), 1-Deoxy-1-L-Leucine-D-Fructose (FRU-ILE), 1-Deoxy-1-L-Leucine-D-Fructose (FRU-LEU), 1-Deoxy-1-L-Asparagine-D-Fructose (FRU-ASN), 1-Deoxy-1-L-Glutamic Acid-D-Fructose (FRU-GLU), 1-Deoxy-1-L-Aminobutyric Acid-D-Fructose (FRU-AMB), 1-Deoxy-1-L-Aspartic Acid-D-Fructose (FRU-AMB) The product contains 70 chemical indicators, including RU-ASP, 1-deoxy-1-L-glutamine-D-fructose (FRU-GLN), 1-deoxy-1-L-glycine-D-fructose (FRU-GLY), 1-deoxy-1-L-histidine-D-fructose (FRU-HIS), 1-deoxy-1-L-threonine-D-fructose (FRU-THR), 1-deoxy-1-L-tyrosine-D-fructose (FRU-TYR), glucosamine (GLU-AN), dichloromethane extract, solanesyl alcohol, and neophytadiene. Including pH value, the product contains a total of 70 chemical indicators.

[0022] The aforementioned chemical components encompass the macro- and semi-trace components of tobacco leaves, excluding cell wall substances. These components undergo thermal migration, thermal decomposition, and thermal transformation into smoke components during cigarette combustion, accounting for 90% of the smoke components by weight. They are precursors to a large number of smoke aroma components. Furthermore, these 69 chemical components exhibit certain correlations with trace aroma components in tobacco leaves. Using these components, information about trace aroma components can be indirectly expressed; for example, the content of vanillin in tobacco leaves is positively correlated with the content of linoleic acid, solanesol, and dichloromethane extract. Therefore, these chemical components are an important material basis for tobacco leaf quality.

[0023] Near-infrared spectroscopy (NIRS) was used to collect the macro- and semi-micro chemical composition data for each tobacco leaf sample. The NIRS method employed a spectral conversion-based prediction approach: the moisture content of all tobacco leaf samples was controlled to 6-8%, and the samples were pulverized to 60-80 mesh. NIRS spectral scanning was then performed on all samples. The true moisture content and first NIRS spectrum of the tobacco leaf samples were obtained. Correction parameters corresponding to the true moisture content were derived from the pre-obtained correspondence between different moisture contents and correction parameters. Based on the correction parameters, the first NIRS spectrum was converted to a second NIRS spectrum at the standard moisture content. The second NIRS spectrum was then input into the NIRS prediction model to output the chemical composition data of the tobacco leaf samples.

[0024] The applicant of this invention has pre-established a prediction model for chemical components in tobacco leaves based on near-infrared spectroscopy ((1) ACSOmega 2022, 7, 43, 38650-38659; (2) A method and device for predicting chemical components of tobacco based on spectral conversion, 202210751386.4; (3) Microchemical Journal 2023, 189, 108522.). Near-infrared spectra of tobacco samples with different moisture contents are pre-obtained. One tobacco sample with a specific moisture content is selected as the standard moisture content tobacco sample. Correction parameters are calculated to convert the near-infrared spectra of other tobacco samples with different moisture contents to the standard moisture content, thus obtaining the correspondence between different moisture contents and the correction parameters. In a practical scenario, the actual moisture content and the first near-infrared spectrum of the tobacco to be tested are obtained. The corresponding correction parameters are extracted from the pre-obtained correspondence based on the actual moisture content. The first near-infrared spectrum under the actual moisture content is converted to the second near-infrared spectrum under the standard moisture content using the correction parameters. The second near-infrared spectrum is input into a pre-established near-infrared prediction model, which outputs the content of chemical components in the tobacco to be tested. This eliminates the need to build a near-infrared prediction model corresponding to the actual moisture content in the laboratory in real time, enabling rapid measurement in real-world scenarios and improving work efficiency. Furthermore, it eliminates the need to build multiple near-infrared prediction models corresponding to different moisture contents in advance, reducing the modeling workload and lowering costs. This lays the foundation for the present invention's method for determining the style of flue-cured tobacco based on 70 chemical components.

[0025] S2. Using chemical composition data, pH value, tobacco leaf part information and origin information as modeling indicators, and sensory indicators of irritating parts as prediction targets, a mapping relationship between modeling indicators and prediction indicators is constructed based on a random forest model. The prediction model is trained and optimized to obtain a prediction model for sensory indicators of irritating parts of tobacco leaves. The sensory indicators of irritating parts include nasal irritation indicators, oral irritation indicators and laryngeal irritation indicators. Training and optimizing the prediction model includes: The method of finding the optimal discrimination threshold is adopted to improve the model prediction accuracy. The criterion for determining the optimal threshold is the F1 score of the model on the validation set. The threshold with the highest F1 score is selected as the optimal discrimination threshold to improve the problem of decreased model prediction accuracy caused by the imbalance of sensory data distribution. The method of finding the optimal threshold was used to improve the modeling problem caused by data imbalance. The effects of the method on the three sensory indicators of the stimulation sites before and after its application are shown in Table 1.

[0026] Table 1

[0027] As can be seen from Table 1, the F1 score predicted by the nasal stimulation index model is improved compared with that before the threshold adjustment, indicating that adjusting the threshold has a certain effect on improving the modeling problem caused by the unbalanced distribution of data.

[0028] Based on the results in Table 1, the parameters of the Random Forest (RF) model for each index model were tuned. Table 2 shows the parameters to be optimized and their values.

[0029] Table 2

[0030] The parameters of the multinomial random forest model in the prediction model were tuned using a lattice search method. The tuning criterion was the F1 score of the model on the validation set. The parameter combination with the highest F1 score was selected as the optimal parameters. After optimization, a prediction model for sensory indicators of tobacco leaf stimulation sites was obtained. The parameters of the multinomial random forest model include the number of decision trees n_estimators, the maximum depth of decision trees max_depth, the maximum number of features in the split max_features, the minimum number of samples required to split internal nodes min_samples_split, and the minimum number of samples contained in a leaf node min_samples_leaf.

[0031] The above parameter combinations total 3*3*3*3*3=243. Each indicator will undergo 243*5=1215 training iterations, for a total of 6*1215=7290 training iterations. The optimal parameter combination will be selected using the 50% average F1 score. Due to the large number of results, Table 3 only shows the optimal parameter combinations for each sensory indicator.

[0032] Table 3

[0033] Compared to before optimization, the model's predictive ability has been improved to some extent. The predictive performance of the model after comprehensive threshold adjustment and parameter tuning is shown in Table 4.

[0034] Table 4

[0035] S3. Based on the optimized sensory index prediction model of tobacco leaf stimulation sites, the sensory index of tobacco leaf stimulation sites is predicted, the key variables in the modeling index are analyzed, and the direction of the key variables' effect on the prediction target is determined.

[0036] Key variables in the analytical modeling metrics include: By ranking the variables in the random forest model by the amount of variance reduction, we calculate the mean of the variance reduction of chemical composition data when splitting across all decision tree nodes. The importance of each feature f is obtained by averaging the mean across all trees. Combined with five-fold cross-validation, the top 10 features f with the highest importance in the five calculations and with at least 4 occurrences are identified as key variables related to the prediction index. The results of the key chemical composition variables are shown in Table 5.

[0037] Table 5

[0038] The results in the table show that Fru-Gln, dichloromethane extract, and Fru-Tyr have a significant impact on sensory indicators at multiple irritation sites.

[0039] Determine the direction of the influence of key variables on the prediction objective, including: For each sensory indicator at a stimulation site, the mean of the key variable was calculated for data groups containing and excluding the sensory indicator at that stimulation site, and the Mann-Whitney U test was used. This test is a non-parametric test used to compare whether the distributions of two groups of samples are significantly different. It has relatively relaxed requirements on the data distribution assumptions and can be used for data that do not meet the requirements of normality or homogeneity of variance. If the mean of the key variable in the data group containing the sensory indicator at the stimulation site is significantly higher than that in the data group without the stimulation feature, then the variable has a positive effect on the sensory indicator at that stimulation site; if the mean is significantly lower than that in the data group without the sensory indicator at the stimulation site, then it has a negative effect. The means of the key variable groups for the sensory indicators at stimulation sites and the test results are shown in Table 6.

[0040] Table 6

[0041] Note: * indicates p-value < 0.05, i.e., significant; ** indicates p-value < 0.01, i.e., highly significant; *** indicates p-value < 0.001, i.e., extremely significant.

[0042] Table 6 shows that the distribution of most key variables differs significantly between the groups with and without stimulation.

[0043] S4: Following the direction of action in step S3, regulate key variables in tobacco leaves to improve tobacco quality.

[0044] For example, in Table 6, neophytadiene has a very significant effect on throat stimulation, and the throat stimulation effect of tobacco leaves can be improved by adjusting the content of neophytadiene.

[0045] To control throat irritation, 0.05 g of food-grade neophytadiene was dissolved in 150 ml of purified water. The solution was then sprayed evenly onto a 1 kg tobacco sample in multiple applications and allowed to air dry. After expert evaluation, the tobacco quality was found to be significantly improved compared to the original sample. This demonstrates that tobacco quality can be improved by controlling key variables in tobacco leaves.

[0046] Device Example 1 This invention provides an electronic device, such as... Figure 2 As shown, it includes: Processor 1020; and, The memory 1010 is configured to store computer-executable instructions, which, when executed, cause the processor 1020 to implement the following steps of a method for analyzing the chemical characteristics of sensory indicators of irritant parts of tobacco leaves: S1. Obtain tobacco leaf samples with different sensory stimulating characteristics, collect the macro and semi-micro chemical composition data, pH value, tobacco leaf part information and origin information for each tobacco leaf sample, and after standardizing each chemical composition data, obtain the modeling dataset, and divide the modeling dataset into training set, validation set and test set according to the proportion. S2. Using chemical composition data, pH value, tobacco leaf part information and origin information as modeling indicators, and sensory indicators of irritating parts as prediction targets, a mapping relationship between modeling indicators and prediction indicators is constructed based on a random forest model. The prediction model is trained and optimized to obtain a prediction model for sensory indicators of irritating parts of tobacco leaves. The sensory indicators of irritating parts include nasal irritation indicators, oral irritation indicators and laryngeal irritation indicators. S3. Based on the optimized sensory index prediction model of tobacco leaf stimulation sites, the sensory index of tobacco leaf stimulation sites is predicted, the key variables in the modeling index are analyzed, and the direction of the key variables' effect on the prediction target is determined.

[0047] Device Example 2 This invention provides a storage medium for storing computer-executable instructions, which, when executed, implement the following steps of a method for analyzing the chemical characteristics of sensory indicators of irritant parts of tobacco leaves: S1. Obtain tobacco leaf samples with different sensory stimulating characteristics, collect the macro and semi-micro chemical composition data, pH value, tobacco leaf part information and origin information for each tobacco leaf sample, and after standardizing each chemical composition data, obtain the modeling dataset, and divide the modeling dataset into training set, validation set and test set according to the proportion. S2. Using chemical composition data, pH value, tobacco leaf part information and origin information as modeling indicators, and sensory indicators of irritating parts as prediction targets, a mapping relationship between modeling indicators and prediction indicators is constructed based on a random forest model. The prediction model is trained and optimized to obtain a prediction model for sensory indicators of irritating parts of tobacco leaves. The sensory indicators of irritating parts include nasal irritation indicators, oral irritation indicators and laryngeal irritation indicators. S3. Based on the optimized sensory index prediction model of tobacco leaf stimulation sites, the sensory index of tobacco leaf stimulation sites is predicted, the key variables in the modeling index are analyzed, and the direction of the key variables' effect on the prediction target is determined.

[0048] The computer-readable storage media in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves, characterized in that, Includes the following steps: S1. Obtain tobacco leaf samples with different sensory stimulating characteristics, collect the macro and semi-micro chemical composition data, pH value, tobacco leaf part information and origin information of each tobacco leaf sample, and after standardizing each chemical composition data, obtain the modeling dataset, and divide the modeling dataset into training set, validation set and test set according to the proportion. S2. Using the chemical composition data, pH value, tobacco leaf part information and origin information as modeling indicators, and the sensory indicators of the irritating parts as prediction targets, a mapping relationship between the modeling indicators and the prediction indicators is constructed based on a random forest model. The prediction model is trained and optimized to obtain a prediction model for the sensory indicators of the irritating parts of tobacco leaves. The sensory indicators of the irritating parts include nasal irritation indicators, oral irritation indicators and laryngeal irritation indicators. S3. Based on the optimized sensory index prediction model of the irritation site of tobacco leaf, the sensory index of the irritation site of the tobacco leaf sample is predicted, the key variables in the modeling index are analyzed, and the direction of the effect of the key variables on the prediction target is determined.

2. The method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves according to claim 1, characterized in that, Near-infrared spectroscopy was used to collect the macro- and semi-micro chemical composition data for each tobacco leaf sample in step S1.

3. The method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves according to claim 2, characterized in that, When acquiring chemical composition data, the near-infrared prediction method adopts a prediction approach based on spectral conversion: the true moisture content and first near-infrared spectrum of the tobacco sample are acquired; the corresponding correction parameters are obtained from the pre-acquired correspondence between different moisture contents and correction parameters; the first near-infrared spectrum is converted into a second near-infrared spectrum under standard moisture content according to the correction parameters; the second near-infrared spectrum is input into the near-infrared prediction model, and the chemical composition data of the tobacco sample is output.

4. The method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves according to claim 1, characterized in that, The training and optimization of the prediction model in step S2 includes: The method of finding the optimal discrimination threshold is adopted to improve the model prediction accuracy. The criterion for determining the optimal threshold is the F1 score of the model on the validation set. The threshold with the highest F1 score is selected as the optimal discrimination threshold to improve the problem of decreased model prediction accuracy caused by the imbalance of sensory data distribution.

5. The method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves according to claim 4, characterized in that, The training and optimization of the prediction model also includes: The parameters of the multinomial random forest model in the prediction model were tuned using a lattice search method. The tuning criterion was the F1 score of the model on the validation set. The parameter combination with the highest F1 score was selected as the optimal parameter. After optimization, a prediction model for sensory indicators of tobacco leaf stimulation sites was obtained. The multinomial random forest model parameters include the number of decision trees n_estimators, the maximum depth of the decision trees max_depth, the maximum number of features to split max_features, the minimum number of samples required to split internal nodes min_samples_split, and the minimum number of samples contained in a leaf node min_samples_leaf.

6. The method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves according to claim 1, characterized in that, The key variables in the analytical modeling metrics in step S3 include: By ranking the variables in the random forest model by the amount of variance reduction, each feature f is calculated, which is the mean of the variance reduction of chemical composition data when splitting across all decision tree nodes. The mean result is averaged over all trees to obtain the importance of each feature f. Combined with five-fold cross-validation, the top 10 features f with the highest importance in the five calculations and appearing at least 4 times are identified as key variables related to the prediction index.

7. The method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves according to claim 1, characterized in that, Determining the direction of the influence of the key variables on the prediction target in step S3 includes: Calculate the mean of the key variable for data groups containing sensory indicators of the stimulation site and data groups without sensory indicators of the stimulation site, and perform a mean difference test. If the mean of the key variable in the data group containing sensory indicators of the stimulation site is significantly higher than that in the data group without sensory indicators of the stimulation site, then the variable has a positive effect on the sensory indicator of the stimulation site; if the mean is significantly lower than that in the data group without sensory indicators of the stimulation site, then it has a negative effect.

8. The method for analyzing the chemical characteristics of sensory indicators of irritating parts of tobacco leaves according to claim 1, characterized in that, It also includes step S4: In accordance with the direction of action in step S3, the key variables in the tobacco leaves are regulated to improve the quality of the tobacco leaves.

9. An electronic device, characterized in that, include: processor; as well as, A memory configured to store computer-executable instructions, which, when executed, cause the processor to implement the steps of the method for analyzing the chemical characteristics of sensory indicators of irritant sites in tobacco leaves as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, Used to store computer-executable instructions, which, when executed, implement the steps of the chemical characteristic analysis method for sensory indicators of irritant sites of tobacco leaves as described in any one of claims 1 to 8.

Images