A method and device for constructing a sensory quality prediction model for tobacco raw materials

Abstract

The present application provides a method and device for constructing a sensory quality prediction model for tobacco leaf raw materials. The method of the present application includes: obtaining tobacco leaf samples and naturally drying the tobacco leaf samples and then crushing them to obtain a sample set; using a prediction model of a near-infrared big data platform based on near-infrared spectroscopy to obtain the content of preset types of tobacco leaf chemical components for each sample in the sample set; obtaining the sensory quality of the sample set; applying grey correlation and principal component regression algorithms to the content of tobacco leaf chemical components and the sensory quality of the sample set, and establishing a sensory index prediction model based on chemical components. The technical solution provided by the present application can more accurately judge the sensory quality of tobacco leaves, avoid subjective differences in the process of manual smoking and judging, and has important application significance for accurately detecting tobacco leaf quality.

Description

Technical Field

[0001] This document relates to the technical field of tobacco leaf sensory evaluation, and in particular to a method and device for constructing a tobacco leaf raw material sensory quality prediction model. Background Art

[0002] The chemical composition of tobacco raw materials is very complex. At present, the evaluation method of tobacco raw materials is mainly based on human sensory evaluation. The core of cigarette design and maintenance is also sensory quality. However, sensory quality evaluation is easily affected by human physical condition and tobacco smoking amount, resulting in fluctuations in sensory quality and low efficiency in obtaining evaluation data. This is not conducive to the transformation of traditional empirical formulation to chemical rational formulation.

[0003] Since the 1970s, the domestic and foreign tobacco industries have revealed some important chemical components related to tobacco leaf quality, which have been widely used in the tobacco industry and have become important reference indicators for tobacco leaf quality evaluation and cigarette product development and maintenance. These tobacco leaf components include total sugar, reducing sugar, total nitrogen, total alkaloids, potassium, chlorine, petroleum ether extract, etc. The derived indicators of these components, such as sugar-alkali ratio, nitrogen-alkali ratio, potassium-chlorine ratio, and Schmuck value (total sugar / protein), also have a significant effect on the sensory quality of smoke. For example, the Schmuck value and sugar-alkali ratio affect the acid-base balance of tobacco leaf smoke, the nitrogen-alkali ratio affects the smoke concentration, and the potassium-chlorine ratio affects the combustibility of tobacco leaves.

[0004] Domestic tobacco companies have conducted many explorations on the relationship between tobacco chemical composition and sensory style quality, but they mostly focus on conventional components and involve limited types. Some studies have found that the nitrogen-alkali ratio, sugar-nitrogen ratio, and sugar-alkali ratio of formula leaf groups have significant effects on sensory quality and style; other studies have pointed out that reducing sugars, aroma components, etc. are also important for sensory quality; other studies have observed that the chemical composition of tobacco leaves in different ecological zones varies greatly, and some chemical indicators are closely related to sensory quality; studies on a specific tobacco area have found that starch, potassium-chlorine ratio, etc. have an impact on sensory quality; recent studies have combined chemical composition with algorithm models to characterize tobacco quality and style, such as using BP neural networks to establish a mathematical model of 8 tobacco leaf components and 8 sensory quality indicators. In summary, current research still focuses on conventional components and their derived indicators. Therefore, it is necessary to provide a sensory prediction model based on multiple chemical components to improve the accuracy of tobacco sensory quality and reduce the subjective errors caused by manual evaluation. Summary of the invention

[0005] The present invention provides a method and device for constructing a sensory quality prediction model for tobacco leaf raw materials, establishes a sensory index prediction model based on multiple chemical components of tobacco leaf raw materials, and introduces the prediction model for predicting tobacco sensory quality based on tobacco chemical components into the tobacco leaf raw material quality identification system, which can more accurately judge the quality of tobacco leaf raw materials and avoid subjective differences in the process of manual smoking and judgment. It has important application significance for accurately detecting the quality of tobacco leaf raw materials and can also reduce the evaluation workload of formula personnel.

[0006] The embodiment of the present invention provides a method for constructing a tobacco leaf raw material sensory quality prediction model, comprising:

[0007] S1. Obtain tobacco leaf samples, dry them naturally, and then crush them to obtain a sample set;

[0008] S2. Using a spectral acquisition instrument to acquire near-infrared spectra of the sample set, and using a prediction model of a near-infrared big data platform based on the near-infrared spectra to acquire the content of chemical components of tobacco leaves of a preset type of each sample in the sample set;

[0009] S3, preparing the sample set into evaluation cigarettes, evaluating the evaluation cigarettes, and scoring them according to a preset number of sensory indicators to obtain a preset number of sensory qualities;

[0010] S4, performing grey correlation analysis on the preset number of sensory qualities and the preset types of tobacco chemical components, and finding chemical indicators with correlations above a first preset percentage as a first indicator set;

[0011] S5. Taking the preset number of sensory qualities as dependent variables and the first indicator set as independent variables, a prediction model is established using the principal component regression analysis method, and principal component regression is performed according to the variance contribution percentage. When the cumulative variance contribution percentage reaches a second preset percentage or above, the final number of principal components is determined.

[0012] The embodiment of the invention provides a device for constructing a tobacco leaf raw material sensory quality prediction model, comprising:

[0013] A sample set acquisition module is used to acquire tobacco leaf samples and naturally dry the tobacco leaf samples and then grind them to acquire a sample set;

[0014] A module for acquiring the content of chemical components of tobacco leaves of a preset type, for acquiring near-infrared spectra of the samples by using a spectral acquisition instrument to acquire spectra of the sample set, and acquiring the content of chemical components of tobacco leaves of a preset type of each sample in the sample set by using a prediction model of a near-infrared big data platform based on the near-infrared spectra;

[0015] A sensory quality module, used for making the sample set into evaluation cigarettes, evaluating the evaluation cigarettes, and scoring them according to a preset number of sensory indicators to obtain a preset number of sensory qualities;

[0016] A correlation analysis module, for performing grey correlation analysis on the preset number of sensory qualities and the preset types of tobacco chemical components, respectively, to find chemical indicators with correlations above a first preset percentage as a first indicator set;

[0017] The regression analysis module is used to establish a prediction model using the principal component regression analysis method with a preset number of sensory qualities as dependent variables and a first indicator set as an independent variable, and to perform principal component regression according to the variance contribution percentage. When the cumulative variance contribution percentage reaches a second preset percentage or above, the final number of principal components is determined.

[0018] An embodiment of the present invention provides an electronic device, including:

[0019] processor; and,

[0020] A memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the steps of the method for constructing the above-mentioned tobacco leaf raw material sensory quality prediction model.

[0021] One or more embodiments of the present specification provide a storage medium for storing computer-executable instructions, which, when executed, implement the steps of the method for constructing the above-mentioned tobacco raw material sensory quality prediction model.

[0022] By adopting the embodiment of the present invention, a sensory index prediction model based on multiple chemical components of tobacco leaf raw materials is established by using the grey correlation + principal component regression algorithm. The prediction model of tobacco chemical components predicting tobacco sensory quality is introduced into the tobacco leaf raw material quality identification system, which is beneficial to the stability of sensory quality and can reduce the workload of formulators in smoking evaluation. It is a supplement and improvement to the traditional evaluation method, making it possible to realize intelligent cigarette formula design and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0024] Figure 1 A flowchart of a method for constructing a tobacco leaf raw material sensory quality prediction model according to an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of a device for constructing a tobacco leaf raw material sensory quality prediction model according to an embodiment of the present invention;

[0026] Figure 3 The error scatter plot and correlation diagram of the observed value and the fitted value of the aroma quantity model of the embodiment of the present invention;

[0027] Figure 4 A schematic diagram of the correlation between the external validation fitting value and the observed value of the aroma quantity model according to an embodiment of the present invention;

[0028] Figure 5 The error scatter plot and correlation diagram of the observed value and the fitted value of the aroma quantity model of the embodiment of the present invention;

[0029] Figure 6 This is a schematic diagram of the correlation between the external validation fitting value and the observed value of the aroma quality model according to an embodiment of the present invention;

[0030] Figure 7 The error scatter plot and correlation plot of the observed value and the fitted value of the permeability model of the embodiment of the present invention;

[0031] Figure 8 It is a schematic diagram of the correlation between the external validation fitting value and the observed value of the permeability model of an embodiment of the present invention;

[0032] Fig. 9 It is an error scatter diagram and a correlation diagram of the observed value and the fitted value of the satisfaction model of the embodiment of the present invention;

[0033] Fig.10 A schematic diagram of the correlation between the external validation fitting value and the observed value of the satisfaction model according to an embodiment of the present invention;

[0034] Fig.11 It is an error scatter plot and correlation plot of the observed value and the fitted value of the salivation sensation model according to an embodiment of the present invention;

[0035] Fig.12 This is a correlation diagram of the external validation fitting value and the observed value of the body fluid production model according to an embodiment of the present invention;

[0036] Fig.13 It is an error scatter plot and correlation diagram of the observed value and the fitted value of the concentration model of the embodiment of the present invention;

[0037] Fig.14 It is a correlation diagram between the fitted value and the observed value of the external verification of the concentration model of the embodiment of the present invention;

[0038] Fig.15 The error scatter diagram and correlation diagram of the observed value and the fitted value of the impurity gas model of the embodiment of the present invention;

[0039] Fig.16 This is a correlation diagram between the fitted value and the observed value of the external verification of the impurity gas model of the embodiment of the present invention;

[0040] Fig.17 It is an error scatter diagram and a correlation diagram of the observed value and the fitted value of the stimulation model of the embodiment of the present invention;

[0041] Fig.18 A correlation diagram of the external validation fitting value and the observed value of the irritation model of an embodiment of the present invention;

[0042] Fig.19 It is an error scatter plot and correlation diagram of the observed value and the fitted value of the aftertaste model of the embodiment of the present invention;

[0043] Fig. 20 It is a correlation diagram between the external validation fitting value and the observed value of the aftertaste model of the embodiment of the present invention;

[0044] Fig.21 It is an error scatter plot and correlation diagram of the original score model observation value and the fitted value in the embodiment of the present invention;

[0045] Fig. 22 This is a schematic diagram of the correlation between the external validation fitting value and the observed value of the original score model of an embodiment of the present invention;

[0046] Fig.23 The error scatter plot and correlation plot of the observed value and the fitted value of the embodiment of the present invention;

[0047] Fig.24 It is a schematic diagram of the correlation between the external validation fitting value and the observed value of the Jintou model in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0049] Method Embodiment

[0050] According to an embodiment of the present invention, a method for constructing a sensory quality prediction model for tobacco leaf raw materials is provided. Figure 1 is a flowchart of a method for constructing a sensory quality prediction model for tobacco leaf raw materials according to an embodiment of the present invention. Figure 1As shown, the method for constructing the tobacco leaf raw material sensory quality prediction model of the embodiment of the present invention specifically includes:

[0051] S1. Obtain tobacco leaf samples, dry them naturally, and then crush them to obtain a sample set;

[0052] In the embodiment of the present invention, the tobacco stems are manually removed from the sample, and the sample is naturally dried to control the moisture content between 6% and 8%. The sample is crushed by a crusher and passed through a 60-mesh sieve. The sample is placed in a ziplock bag and sealed after exhausting all the air.

[0053] S2. Using a spectral acquisition instrument to acquire near-infrared spectra of the sample set, and using a prediction model of a near-infrared big data platform based on the near-infrared spectra to acquire the content of chemical components of tobacco leaves of a preset type of each sample in the sample set;

[0054] In one embodiment of the present invention, the ambient temperature is set to 22±2° C., the relative humidity is set to 60±5%, and the sample spectrum data is collected.

[0055] In one embodiment of the present invention, the spectrum acquisition instrument parameters are set as follows: the scanning range is 10000cm -1 ~4000cm -1 , resolution 8cm -1 ; The spectrum was scanned 64 times; the light spot of the integrating sphere fell within 2 / 3 of the center of the sampling cup, the near-infrared spectra of all samples were collected twice (with re-sampling each time), and the two spectra should pass the consistency test (spectral similarity greater than 0.9999), and the average spectrum was taken as the near-infrared spectrum of the sample.

[0056] Based on the near-infrared spectrum of tobacco leaves and the prediction model of the tobacco near-infrared big data platform, the conventional chemical components of tobacco leaves, dichloromethane extract, solanesol, neophytadiene, pH, inorganic anions and cations, polyphenols, non-volatile acids and higher fatty acids, amino acids, Amadori compounds and other chemical components were obtained.

[0057] The tobacco leaves are cut into shreds after the main veins are removed manually, and rolled into smoking cigarettes without filter rods. The moisture content of the cigarettes is adjusted according to the requirements of GB / T16447-2004 "Atmospheric Environment for Conditioning and Testing of Tobacco and Tobacco Products" for sensory evaluation. A number of sensory evaluation professionals form an evaluation team, using a percentage system. The evaluation indicators include aroma quality, aroma volume, permeability, satisfaction, salivation, concentration, miscellaneous smell, irritation, aftertaste, and the original total score of sensory quality, a total of 11 indicators (of which the strength is not included in the total score). The arithmetic mean of each indicator is calculated, which is the result of each indicator of the sample.

[0058] The original score = aroma quality + aroma quantity + permeability + satisfaction + salivation + concentration + miscellaneous smell + stimulation + aftertaste.

[0059] S3, performing grey correlation analysis on the preset number of sensory indicators and the preset types of tobacco chemical components, and finding chemical indicators with correlations above a first preset percentage as the first indicator set;

[0060] Grey correlation analysis was performed on the sensory quality indicators and the 72 chemical indicators in Table 1 above to find out the chemical indicators with a correlation of 80% or above. The test data were processed using standardized data.

[0061] S4. Using a preset number of sensory indicators as dependent variables and the first indicator set as independent variables, respectively, using principal component regression analysis to establish prediction models for the sensory indicators, wherein the principal component regression analysis method specifically includes: performing principal component regression according to the variance contribution percentage in the first indicator set, and when the cumulative variance contribution percentage reaches a second preset percentage or above, determining the final number of principal components related to the sensory indicators.

[0062] The prediction model of the sensory index specifically includes:

[0063]

[0064] Among them, A represents the regression coefficient, n represents the number of chemical indicators entering the sensory prediction model, and B i represents the i-th chemical index entering the sensory prediction model, a i Represents the regression coefficient corresponding to the i-th chemical indicator.

[0065] The embodiment of the present invention analyzes 72 chemical indicators and finally constructs an aroma quality prediction model, an aroma quantity prediction model, a permeability prediction model, a satisfaction prediction model, a salivation prediction model, a concentration prediction model, a miscellaneous smell prediction model, an irritation prediction model, an aftertaste prediction model, a sensory original score prediction model and a strength prediction model.

[0066] The following is a specific example to illustrate the specific process of using grey correlation analysis + principal component regression analysis to analyze 72 chemical indicators of tobacco leaves and then realize tobacco leaf sensory quality modeling. The specific 72 chemical components are as follows:

[0067] Total plant alkaloids (X1), reducing sugar (X2), total sugar (X3), total nitrogen (X4), potassium ion (X5), chloride ion (X6), pH (X7), starch (X8), dichloromethane extract (X9), solanesol (X10), sodium pyrophosphate (X11), sodium pyrophosphate (X12), sodium pyrophosphate (X13), sodium pyrophosphate (X14), sodium pyrophosphate (X15), sodium pyrophosphate (X16), sodium pyrophosphate (X17), sodium pyrophosphate (X18), sodium pyrophosphate (X19), sodium pyrophosphate (X20), sodium pyrophosphate (X21), sodium pyrophosphate (X22), sodium pyrophosphate (X23), sodium pyrophosphate (X24), sodium pyrophosphate (X25), sodium pyrophosphate (X26), sodium pyrophosphate (X27), sodium pyrophosphate (X28), sodium pyro 10 ), sulfate (X 11 ), phosphate (X 12 ), Mg2+ (X 13 ), Ca 2+ (X 14 ), neochlorogenic acid (X 15 ), chlorogenic acid (X 16 ), cryptochlorogenic acid (X 17 ), Scopolamine (X 18 ), rutin (X 19 ), lactic acid (X 20 ), oxalic acid (X 21 ), malonic acid (X 22 ), levulinic acid (X 23 ), succinic acid (X 24 ), malic acid (X 25 ), citric acid (X 26 ), vanillic acid (X 27 ), tetradecanoic acid (X 28 ), hexadecanoic acid (X 29 ), linoleic acid (X 30 ), oleic acid + linolenic acid (X 31 ), octadecanoic acid (X 32 ), eicosanoic acid (X 33 ), aspartic acid (X 34 ), threonine (X 35 ), serine (X 36 ), asparagine (X 37 ), glutamate (X 38 ), glutamine (X 39 ), glycine (X 40 ), alanine (X 41 ), valine (X 42 ), cystine (X 43 ), methionine (X 44 ), isoleucine (X 45 ), leucine (X 46 ), tyrosine (X 47 ), phenylalanine (X 48 ), 4-aminobutyric acid (X 49 ), lysine (X 50 ), histidine (X 51 ), tryptophan (X 52 ), arginine (X 53 ), proline (X 54 ), glucosamine (X 55 ), 1-deoxy-1-L-aminobutyric acid-D-fructose (X 56 ), 1-deoxy-1-L-histidine-D-fructose (X 57 ), 1-deoxy-1-L-proline-D-fructose (X 58), 1-deoxy-1-L-valine-D-fructose (X 59 ), 1-deoxy-1-L-threonine-D-fructose (X 60 ), 1-deoxy-1-L-glycine-D-fructose (X 61 ), 1-deoxy-1-L-alanine-D-fructose (X 62 ), 1-deoxy-1-L-asparagine-D-fructose (X 63 ), 1-deoxy-1-L-aspartate-D-fructose (X 64 ), 1-deoxy-1-L-glutamine-D-fructose (X 65 ), 1-deoxy-1-L-glutamate-D-fructose (X 66 ), 1-deoxy-1-L-isoleucine-D-fructose (X 67 ), 1-deoxy-1-L-leucine-D-fructose (X 68 ), 1-deoxy-1-L-tyrosine-D-fructose (X 69 ), 1-deoxy-1-L-phenylalanine-D-fructose (X 70 ), 1-deoxy-1-L-tryptophan-D-fructose (X 71 ), neophytadiene (X 72 ).

[0068] 1. Aroma volume modeling process:

[0069] First, a grey correlation analysis was performed. Under the DPS7.05 statistical software operating environment, a grey correlation analysis was performed on the aroma volume and the above 72 chemical indicators. The correlation between the aroma volume and each chemical indicator with a correlation of 80% or above was considered to be good, so as to achieve the purpose of modeling and dimensionality reduction, and the chemical indicators with a correlation of more than 80% were taken as the first indicator set.

[0070] The test data were processed using standardized data, and the analysis parameters were set to min = 0 and resolution coefficient = 0.5. The chemical indices with gray correlations of aroma volume and chemical indices above 80% are shown in Table 1.

[0071] As shown in Table 1, the correlation between aroma volume and 31 chemical indicators is above 80%, and the correlation is sorted from large to small in the table.

[0072] Table 1 Grey correlation system analysis of aroma quantity and factors

[0073]

[0074]

[0075] Then the model is established by principal component regression method:

[0076] In the operating environment of DPS7.05 statistical software, the prediction model was established by principal component regression analysis method, with aroma volume as the dependent variable and 31 chemical indicators with grey correlation degree above 80% as independent variables.

[0077] Principal component regression was performed according to the variance contribution percentage. When the cumulative contribution percentage was 90.0925%, the final number of principal components was 13. The results of the principal component analysis of the independent variables are shown in Table 2.

[0078] The principal component regression model and correlation analysis results of the dependent and independent variables are shown in Table 3. Table 3 shows that the model and its variables are both significant or extremely significant, with a multiple correlation coefficient of 0.8549 and an adjusted correlation coefficient of 0.8483, indicating that the established prediction model is significant.

[0079] Depend on Figure 3 The error scatter plot on the left shows that the accuracy of the aroma volume prediction model is relatively good. Among them, there are 6 groups with fitting errors of ±3 points or more, accounting for 0.76%, 16 groups with ±2 to ±3 points, accounting for 1.80%, 809 groups with ±1.5 points or less, accounting for 90%, and 716 groups with ±1 points or less, accounting for 80.54%. Figure 3 The correlation diagram between the observed values ​​and the fitted values ​​in the right figure shows that there is a significant linear correlation between the model fitted values ​​and the observed values, and the correlation coefficient R2 is 0.8354.

[0080] The chemical indicators of aroma prediction model are: reducing sugar, potassium ion, chloride ion, pH, solanesol, phosphate, Mg 2+ , Ca 2+ , chlorogenic acid, rutin, malonic acid, vanillic acid, and linoleic acid.

[0081] The main component regression prediction model of aroma volume finally established is:

[0082] Y=24.0386-0.0212X2-0.4974X5+0.0276X6-0.9828X7+0.0206X 10 -0.5525X1

[0083] 2-2.7994X 13 -0.7685X 14 -0.1012X 16 +0.0150X 19 +0.0442X 22 -9.0943X 27 -0.2748X 30 ;

[0084] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0085] Table 2 Results of principal component analysis of independent variables

[0086] Serial number Eigenvalue Contribution percentage / % Cumulative contribution percentage / % 1 2.7314 30.1842 30.1842 2 2.5539 11.9831 42.1673 3 2.1937 9.5763 51.7436 4 1.9760 7.0167 58.7603 5 1.6671 6.4217 65.182 6 1.4129 4.3178 69.4998 7 1.2814 4.0256 73.5254 8 1.0115 3.7017 77.2271 9 0.8378 3.3679 80.595 10 0.6917 3.1127 83.7077 11 0.5238 2.4718 86.1795 12 0.3196 2.1639 88.3434 13 0.2273 1.7491 90.0925

[0087] Table 3. Relevant statistical parameters of principal component regression model

[0088]

[0089] 50 groups of tobacco leaf samples of different grades from different regions other than those used for modeling were randomly selected for external validation of the model to examine the prediction ability of the model by verifying the prediction standard deviation (RMSEP) and average prediction relative deviation of the samples; the prediction results of the validation samples are shown in Table 4:

[0090] Table 4 Model prediction results

[0091] Model RMSEP Relative deviation / % <![CDATA[Relevance (R 2 )]]> Aroma volume (Y) 1.07 9.7 0.85

[0092] Figure 4 This is a schematic diagram of the correlation between the fitted value and the observed value of the external validation of the aroma quantity model. Figure 4 It can be seen that the prediction standard deviation (RMSEP) of the external validation model is 1.07, the average prediction relative deviation is 9.7%, and the correlation coefficient R 2 It is 0.85, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual value, and the average prediction relative deviation is less than 10%.

[0093] The above-mentioned aroma quantity modeling method is also used to establish a prediction model for the ten dependent variable parameters, including aroma quality, transparency, satisfaction, salivation, concentration, miscellaneous smell, irritation, aftertaste, original score, and strength. The chemical indicators entering the model, the prediction fitting error, and the linear correlation between the observed value and the fitted value are now described separately according to the type of dependent variable.

[0094] 2. Fragrance quality modeling process

[0095] The chemical indicators that enter the aroma quality prediction model are: oleic acid + linolenic acid, levulinic acid, reducing sugar, lactic acid, total sugar, hexadecanoic acid, neophytadiene, octadecanoic acid, potassium ion, total plant alkaloids, phosphate, rutin, succinic acid, 1-deoxy-1-L-aminobutyric acid-D-fructose, and proline.

[0096] The final aroma quality principal component regression prediction model established is:

[0097] Y=36.0700+0.5032X 31 +0.8165X 23- 0.2681X2-8.6068X 20-0.5788X3-14.2753

[0098] X 29 +3.4772X 72 +14.4923X 32 -1.9573X5-0.1972X1+0.2912X 12 +0.1544X 19 +10.5567X 24 -0.0009X 56 -0.0002X 54 ;

[0099] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0100] Depend on Figure 5 The error scatter plot on the left shows that the accuracy of the aroma quality prediction model is relatively good, with a prediction standard deviation (RMSEP) of 1.0123. Among them, there are 4 groups with fitting errors of ±3 points or more, accounting for 0.45%, 34 groups with ±2 to ±3 points, accounting for 3.82%, 796 groups with ±1.5 points or less, accounting for 88.53%, and 735 groups with ±1 or less, accounting for 82.68%. Figure 5 The correlation diagram on the right shows that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8143.

[0101] Figure 6 This is the correlation diagram of the external validation of the aroma quality prediction model; the results show that the prediction standard deviation (RMSEP) of the external validation model is 0.92, the average prediction relative deviation is 8.1%, and the correlation coefficient R 2 It is 0.8359, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual values, and the average prediction relative deviation is less than 10%.

[0102] 3. Transparency Modeling Process

[0103] The chemical indicators that enter the permeability prediction model are: hexadecanoic acid, octadecanoic acid, 1-deoxy-1-L-aminobutyric acid-D-fructose, phosphate, oxalic acid, linoleic acid, reducing sugar, chlorogenic acid, levulinic acid, neophytadiene, total sugar, 1-deoxy-1-L-valine-D-fructose, cryptochlorogenic acid, succinic acid, alanine, lactic acid, potassium ion, 1-deoxy-1-L-phenylalanine-D-fructose, proline, cystine, 1-deoxy-1-L-threonine-D-fructose.

[0104] The principal component regression prediction model of permeability that was finally established is:

[0105] Y=20.4149-7.9305X 29 +14.3888X 32 -0.0006X 56 +0.2909X 12 -0.0606X 21 +0.02

[0106] 42X 30 -0.0820X2+0.0626X 16 +0.2759X 23 +1.2534X 72 -0.1528X3+0.0016X 59 -0.2049X 17 +4.9435X 24 -0.0025X 41 -4.7728X 20 -0.7406X5-0.0010X 70 -0.0000087X 54 +0.0021X 55 +0.0356X 60 ;

[0107] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0108] Depend on Figure 7 The error scatter plot on the left shows that the accuracy of the permeability prediction model is relatively good, with a prediction standard deviation (RMSEP) of 0.5069. Among them, there are 10 groups with fitting errors of ±1 point or more, accounting for 1.12%, 134 groups between ±0.5 and ±1 points, accounting for 15.07%, and 765 groups within ±0.5 points, accounting for 86.05%.

[0109] Depend on Figure 7 The right correlation diagram shows that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8385.

[0110] Figure 8 The results of the external validation correlation diagram showed that the prediction standard deviation (RMSEP) of the external validation model was 0.71, the average prediction relative deviation was 6.8%, and the correlation coefficient R 2 It is 0.9948, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual values, and the average prediction relative deviation is significantly lower than 10%.

[0111] 4. Satisfaction Modeling Process

[0112] The chemical indicators that entered the satisfaction prediction model were: linoleic acid, total alkaloids, neophytadiene, vanillic acid, 1-deoxy-1-L-alanine-D-fructose, dichloromethane extract, starch, 1-deoxy-1-L-proline-D-fructose, total nitrogen, scopoletin, and rutin. The final satisfaction principal component regression prediction model established was:

[0113] Y=6.2724+0.4072X 30 +0.7487X1+1.0048X 72 -7.5127X 27 -0.0002X 62 -0.0699

[0114] X9+0.0486X8+0.00009X 58 -1.2468X4+1.6572X 18 +0.0148X 19 ;

[0115] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0116] Depend on Fig. 9 The error scatter plot shows that the accuracy of the satisfaction prediction model is relatively good, with a prediction standard deviation (RMSEP) of 1.0553. Among them, there are 8 groups with a fitting error of ±1 point or more, accounting for 0.9%, 122 groups between ±0.5 and ±1 points, accounting for 13.72%, and 775 groups within ±0.5 points, accounting for 87.18%.

[0117] Depend on Fig. 9 The correlation diagram shows that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8955.

[0118] Fig.10 The results show that the prediction standard deviation (RMSEP) of the external validation model is 1.11, and the average prediction relative deviation and correlation coefficient R 2 They are 9.8% and 0.8483 respectively, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model's prediction results are consistent with the actual values, and the average prediction relative deviation is slightly less than 10%.

[0119] 5. Modeling process of the sensation of salivation

[0120] The chemical indicators that entered the model in the prediction model of the sense of body fluid production are: 1-deoxy-1-L-proline-D-fructose, 1-deoxy-1-L-alanine-D-fructose, starch, levulinic acid, reducing sugar, hexadecanoic acid, 1-deoxy-1-L-tryptophan-D-fructose, rutin, total plant alkaloids, neophytadiene, octadecanoic acid, 1-deoxy-1-L-aspartic acid-D-fructose. The principal component regression prediction model of the sense of body fluid production that was finally established is:

[0121] Y=10.6238+0.000044X 58 -0.0037X 62 +0.0710X8+0.3869X 23 -0.1531X2-7.51

[0122] 36X 29 -0.0023X 71 +0.1328X 19 +1.0815X1+1.8629X 72 +18.6314X 32 +0.0010X 64 ;

[0123] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0124] Depend on Fig.11 The error scatter plot shows that the prediction model for salivation is relatively accurate, with a prediction standard deviation (RMSEP) of 0.8607. Among them, the fitting error was ±2 points or more in 2 groups, accounting for 0.22%, ±1 point or more in 61 groups, accounting for 6.86%, ±0.5 to ±1 points in 187 groups, accounting for 21.03%, ±1 point or less in 826 groups, accounting for 92.91%, and ±0.5 points or less in 639 groups, accounting for 71.88%.

[0125] Depend on Fig.11 The correlation diagram shows that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8988.

[0126] For external validation relevance, see Fig.12 The results showed that the prediction standard deviation (RMSEP) of the external validation model was 0.81, the average prediction relative deviation was 8.8%, and the correlation coefficient R 2 It is 0.9295, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual values, and the average prediction relative deviation is less than 10%.

[0127] 6. Concentration modeling process

[0128] The chemical indicators that entered the concentration prediction model were: total alkaloids, linoleic acid, neophytadiene, chloride ion, 1-deoxy-1-L-proline-D-fructose, starch, proline, phosphate, 1-deoxy-1-L-tryptophan-D-fructose, 1-deoxy-1-L-alanine-D-fructose, rutin, glucosamine, and arginine.

[0129] The concentration principal component regression prediction model finally established is:

[0130] Y=5.7928+0.5797X1+0.9684X 30 +0.9182X 72 +0.0229X6+0.0001X 58 +0.0499

[0131] X8+0.000023X 54 +0.2396X 12 -+0.0012X 71 -0.0018X 62 +0.0782X 19 +0.0020X 55 -0.0633X 53 ;

[0132] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0133] Depend on Fig.13 It can be seen that the concentration prediction model has relatively good accuracy, with a prediction standard deviation (RMSEP) of 0.9649. Among them, the fitting error was ±2 points or more for 4 groups, accounting for 0.45%, ±1 point or more for 25 groups, accounting for 2.81%, ±0.5 to ±1 points for 128 groups, accounting for 14.4%, ±1 point or less for 860 groups, accounting for 96.74%, and ±0.5 points or less for 732 groups, accounting for 82.34%.

[0134] Depend on Fig.13 It can be seen that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8827.

[0135] Fig.14 This is a schematic diagram of the correlation between the fitted value and the observed value of the external validation of the concentration model. Fig.14 The results showed that the prediction standard deviation (RMSEP) of the external validation model was 0.99, the average prediction relative deviation was 9.1%, and the correlation coefficient R 2It is 0.8905, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual values, and the average prediction relative deviation is slightly less than 10%.

[0136] 7. Impurity modeling process

[0137] The chemical indicators entering the impurity gas prediction model are: 1-deoxy-1-L-proline-D-fructose, 1-deoxy-1-L-alanine-D-fructose, starch, levulinic acid, total sugar, hexadecanoic acid, lactic acid, octadecanoic acid, neophytadiene, chloride ion, proline, rutin, 1-deoxy-1-L-aminobutyric acid-D-fructose, 1-deoxy-1-L-histidine-D-fructose, solanesol, and methionine.

[0138] The main component regression prediction model of impurity gas finally established is:

[0139] Y=16.0431+0.0001X 58 -0.0013X 62 +0.1023X8+0.2070X 23 -0.1807X3-4.3782

[0140] X 29 -1.6954X 20 +5.7084X 32 +0.9651X 72 -0.0161X6-0.0002X 54 +0.0514X 19 -0.0005X 56 +0.0112X 57 +0.0343X 10 +0.0519X 44 ;

[0141] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0142] Depend on Fig.15 It can be seen that the accuracy of the impurity gas prediction model is relatively good, with a prediction standard deviation (RMSEP) of 1.0265. Among them, the fitting error is ±1 point or above for 13 groups, accounting for 1.46%, between ±0.5 and ±1 points for 73 groups, accounting for 8.21%, and within ±0.5 points for 803 groups, accounting for 90.33%.

[0143] Depend on Fig.15 It can be seen that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8419.

[0144] Fig.16The correlation diagram of the external validation fitting value and the observed value of the impurity model shows that the prediction standard deviation (RMSEP) of the external validation model is 1.03, the average prediction relative deviation is 9.4%, and the correlation coefficient R 2 It is 0.8766, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model's prediction results are consistent with the actual values, and the average prediction relative deviation is less than 10%.

[0145] 8. Stimulating Modeling Process

[0146] The chemical indicators that enter the irritation prediction model are: levulinic acid, 1-deoxy-1-L-alanine-D-fructose, reducing sugar, hexadecanoic acid, lactic acid, Mg 2+ , rutin, neophytadiene, 1-deoxy-1-L-histidine-D-fructose, 1-deoxy-1-L-aminobutyric acid-D-fructose, 1-deoxy-1-L-glutamic acid-D-fructose, 1-deoxy-1-L-threonine-D-fructose, serine, methionine, linoleic acid, 1-deoxy-1-L-glutamine-D-fructose. The principal component regression prediction model of pungency established in the end was:

[0147] Y=18.0623+0.0719X 23 -0.0009X 62 +0.0354X2-3.3573X 29 -3.5464X 20 -0.1371

[0148] X 13 +0.0507X 19 +1.2641X 72 +0.0082X 57 -0.0004X 56 +0.0012X 66 -0.0227X 60 +0.0009X 36 +0.0392X 44 -0.5796X 30 -0.0015X 65 ;

[0149] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0150] Depend on Fig.17As shown in the left figure, the accuracy of the stimulus prediction model is relatively good, with a prediction standard deviation (RMSEP) of 1.4280. Among them, 22 groups had a fitting error of ±1 point or more, accounting for 2.47%, 149 groups had a fitting error between ±0.5 and ±1 point, accounting for 16.76%, and 718 groups had a fitting error within ±0.5 point or less, accounting for 80.76%.

[0151] Depend on Fig.17 The correlation diagram shows that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.7789.

[0152] Fig.18 This is a correlation diagram between the fitted values ​​and observed values ​​of the external validation of the stimulus model. The results show that the prediction standard deviation (RMSEP) of the external validation model is 1.71, the average prediction relative deviation is 10.4%, and the correlation coefficient R2 is 0.7963, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model's prediction results are consistent with the actual values, and the average prediction relative deviation is close to 10%.

[0153] 9. Aftertaste Modeling Process

[0154] The chemical indicators entering the aftertaste prediction model are: 1-deoxy-1-L-alanine-D-fructose, 1-deoxy-1-L-proline-D-fructose, reducing sugar, hexadecanoic acid, lactic acid, 1-deoxy-1-L-histidine-D-fructose, serine, 1-deoxy-1-L-glutamate-D-fructose, neophytadiene, linoleic acid, 1-deoxy-1-L-glutamine-D-fructose, Mg 2+ , 1-deoxy-1-L-glycine-D-fructose, malic acid.

[0155] The final established aftertaste principal component regression prediction model is:

[0156] Y=24.1268-0.0008X 62 +0.000088X 58 +0.0657X2-4.3180X 29 -3.8292X 20 +0.00

[0157] 77X 57 +0.0020X 36 +0.0017X 66 +1.0910X 72 -0.9075X 30 -0.0021X 65 -0.7027X 13 -0.0538X 61 -0.0129X 25;

[0158] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0159] Depend on Fig.19 It can be seen that the aftertaste prediction model has relatively good accuracy, and the prediction standard deviation (RMSEP) is 1.2696. Among them, there are 12 groups with fitting errors of ±1 point or above, accounting for 1.35%, 165 groups between ±0.5 and ±1 points, accounting for 18.56%, and 712 groups within ±0.5 points, accounting for 80.09%.

[0160] Depend on Fig. 20 The correlation diagram shows that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8172. Fig. 20 The correlation diagram of the external validation fitting value and the observed value of the flavor model is shown in Figure 2. The results show that the prediction standard deviation (RMSEP) of the external validation model is 1.23, the average prediction relative deviation is 9.9%, and the correlation coefficient R 2 It is 0.8325, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual values, and the average prediction relative deviation is slightly less than 10%.

[0161] 10. Sensory raw score modeling process:

[0162] The chemical indicators that entered the model of the original score prediction model were: starch, levulinic acid, linoleic acid, reducing sugar, oleic acid + linolenic acid, total sugar, lactic acid, total alkaloids, rutin, 1-deoxy-1-L-tryptophan-D-fructose, neophytadiene, octadecanoic acid, 1-deoxy-1-L-histidine-D-fructose, 1-deoxy-1-L-aspartate-D-fructose, chlorogenic acid, solanesol, phosphate, 1-deoxy-1 -L-aminobutyric acid-D-fructose, neochlorogenic acid, potassium ion, dichloromethane extract, 1-deoxy-1-L-glutamic acid-D-fructose, serine, 1-deoxy-1-L-valine-D-fructose, glucosamine, 1-deoxy-1-L-glutamine-D-fructose, tryptophan, isoleucine, 1-deoxy-1-L-tyrosine-D-fructose, 1-deoxy-1-L-isoleucine-D-fructose, total nitrogen.

[0163] The sensory original principal component regression prediction model finally established is:

[0164] Y=178.1866-2.1585X8+0.0213X 23 -10.6884X 30 +0.5065X2+0.3434X 31 -0.14

[0165] 43X3-79.7779X 20 -6.0158X1+0.2392X 19 +0.0439X 71 +20.4080X 72 +10.3136X 32 +0.1759X 57 +0.0209X 64 +0.8887X 16 -1.1027X 10 -1.6046X 12 -0.0110X 56 -6.0739X 15 -8.2938X5+5.1613X9+0.0254X 66 +0.0059X 36 -0.0367X 59 +0.0193X 55 -0.0287X 65 +0.0484X 52 -2.5506X 45 -0.3733X 69 +0.6112X 67 -2.8237X4;

[0166] The corresponding independent variables in the model can be specifically corresponded to the 72 chemical components disclosed above.

[0167] Depend on Fig.21 As can be seen from the left figure, the prediction model of the original total sensory score has a relatively good accuracy, with a prediction standard deviation (RMSEP) of 1.0631. Among them, there are 9 groups with a fitting error of ±5 points or more, accounting for 1.01%, 163 groups between ±3 and ±5 points, accounting for 18.34%, and 717 groups within ±3 points, accounting for 80.65%. Fig.21 The right figure shows that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8822. Fig. 22 The external validation correlation diagram of the original sensory scores. The results show that the prediction standard deviation (RMSEP) of the external validation model is 1.01, the average prediction relative deviation is 9.1%, and the correlation coefficient R 2 It is 0.8421, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual values, and the average prediction relative deviation is less than 10%.

[0168] 11. Momentum modeling process

[0169] The chemical indicators that enter the Jintou prediction model are: total alkaloids, linoleic acid, solanesol, total nitrogen, glycine, chloride ion, 1-deoxy-1-L-glycine-D-fructose, glucosamine, scopoletin, proline, 1-deoxy-1-L-asparagine-D-fructose, 1-deoxy-1-L-tryptophan-D-fructose, glutamic acid, and potassium ion.

[0170] The main component regression prediction model finally established is:

[0171] Y=3.6554+0.5372X1+0.4321X 30 -0.0247X 10 -0.5203X4+0.0778X 40 -0.0126X

[0172] 6-0.0252X 61 +0.0013X 55 +0.7955X 18 +0.000063X 54 +0.0001X 63 -0.0004X 71 -0.0027X 38 +0.0986X5;

[0173] The independent variables in the model can be specifically referred to the 72 chemical components disclosed above. Fig.23 It can be seen that the accuracy of the momentum prediction model is relatively good, with a prediction standard deviation (RMSEP) of 1.1644. Among them, there are 2 groups with a fitting error of ±1 point or more, accounting for 0.22%, 107 groups between ±0.5 and ±1 points, accounting for 12.04%, and 780 groups within ±0.5 points, accounting for 87.74%. Fig.23 It can be seen that there is a significant linear correlation between the model fitting value and the observed value, and the correlation coefficient R 2 It is 0.8203. Fig.24 The results of the external validation diagram showed that the prediction standard deviation (RMSEP) of the external validation model was 1.11, the average prediction relative deviation was 9.7%, and the correlation coefficient R 2 It is 0.8405, indicating that there is a significant linear correlation between the model prediction value and the standard value; the model prediction results are consistent with the actual values, and the average prediction relative deviation is slightly less than 10%.

[0174] The embodiments of the present invention have the following beneficial effects:

[0175] By adopting the embodiment of the present invention, a sensory index prediction model based on multiple chemical components of tobacco leaf raw materials is established by using the grey correlation + principal component regression algorithm. The prediction model of tobacco sensory quality predicted by tobacco chemical components is introduced into the tobacco leaf raw material quality identification system, which is beneficial to the stability of sensory quality and can reduce the workload of formulators in smoking evaluation. It is a supplement and improvement to the traditional evaluation method, making it possible to realize intelligent cigarette formula design and maintenance.

[0176] Device Example 1

[0177] According to an embodiment of the present invention, a device for constructing a tobacco leaf raw material sensory quality prediction model is provided. Figure 2 Schematic diagram of a device for constructing a sensory quality prediction model for tobacco leaf raw materials according to an embodiment of the present invention. Figure 2 As shown, the device for constructing a tobacco leaf raw material sensory quality prediction model according to an embodiment of the present invention specifically includes:

[0178] A sample set acquisition module is used to acquire tobacco leaf samples and naturally dry the tobacco leaf samples and then grind them to acquire a sample set;

[0179] A module for acquiring the content of chemical components of tobacco leaves of a preset type, for acquiring near-infrared spectra of the samples by using a spectral acquisition instrument to acquire spectra of the sample set, and acquiring the content of chemical components of tobacco leaves of a preset type of each sample in the sample set by using a prediction model of a near-infrared big data platform based on the near-infrared spectra;

[0180] A sensory quality module, used for making the sample set into evaluation cigarettes, evaluating the evaluation cigarettes, and scoring them according to a preset number of sensory indicators to obtain a preset number of sensory qualities;

[0181] A correlation analysis module, for performing grey correlation analysis on the preset number of sensory qualities and the preset types of tobacco chemical components, respectively, to find chemical indicators with correlations above a first preset percentage as a first indicator set;

[0182] The regression analysis module is used to establish a prediction model using the principal component regression analysis method with a preset number of sensory qualities as dependent variables and a first indicator set as an independent variable, and to perform principal component regression according to the variance contribution percentage. When the cumulative variance contribution percentage reaches a second preset percentage or above, the final number of principal components is determined.

[0183] Device Example 2

[0184] According to an embodiment of the present invention, an electronic device is provided, including:

[0185] processor; and,

[0186] A memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the steps as described in the above method embodiment.

[0187] Device Example 3

[0188] According to an embodiment of the present invention, a storage medium is provided for storing computer executable instructions, and the computer executable instructions, when executed, implement the steps described in the above method embodiment.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements 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 constructing a sensory quality prediction model for tobacco leaf raw materials, characterized in that include: S1. Obtain tobacco leaf samples, dry them naturally, and then crush them to obtain a sample set; S2. Using a spectral acquisition instrument to acquire near-infrared spectra of the sample set, and using a prediction model of a near-infrared big data platform based on the near-infrared spectra to acquire the content of chemical components of tobacco leaves of a preset type of each sample in the sample set; S3, performing grey correlation analysis on the preset number of sensory indicators and the preset types of tobacco chemical components, and finding chemical indicators with correlations above a first preset percentage as the first indicator set; S4, respectively using a preset number of sensory indicators as dependent variables and the first indicator set as independent variables, using principal component regression analysis to establish prediction models for the sensory indicators, and predicting the sensory quality of the tobacco raw material according to the prediction models for the sensory indicators; The principal component regression analysis method specifically includes: performing principal component regression according to the variance contribution percentage in the first indicator set, and when the accumulated variance contribution percentage reaches a second preset percentage or above, determining the final number of principal components related to the sensory index.

2. The method according to claim 1, characterized in that The preset number of sensory indicators specifically include: aroma quality, aroma quantity, permeability, satisfaction, salivation, concentration, miscellaneous smell, irritation, aftertaste, original score and strength.

3. The method according to claim 2, characterized in that The original score is the sum of aroma quality, aroma quantity, permeability, satisfaction, salivation, concentration, miscellaneous smell, irritation and aftertaste.

4. The method according to claim 1, characterized in that: The S1 specifically includes: The moisture content of the tobacco leaf samples is controlled within a preset range, the tobacco leaf samples are crushed by a crusher and passed through a 60-mesh sieve, and the tobacco leaf samples are placed in a ziplock bag and sealed in a vacuum state.

5. The method according to claim 1, characterized in that The near infrared spectrum of the sample is obtained by: The spectral data of the sample set is collected according to preset spectral collection instrument parameters under preset temperature and preset humidity conditions.

6. The method according to claim 5, characterized in that The obtaining of the near infrared spectrum of the sample further comprises: The near-infrared spectra of all samples in the sample set were collected twice, and after the spectra collected twice passed the consistency test, the average spectrum was taken as the near-infrared spectrum of the sample.

7. The method according to claim 2, characterized in that The prediction model of the sensory index specifically includes: Among them, A represents the regression coefficient, n represents the number of chemical indicators entering the sensory prediction model, and B i represents the i-th chemical index entering the sensory prediction model, a i Represents the regression coefficient corresponding to the i-th chemical indicator.

8. A device for constructing a tobacco leaf raw material sensory quality prediction model, characterized in that: include: A sample set acquisition module is used to acquire tobacco leaf samples and naturally dry the tobacco leaf samples and then grind them to obtain a sample set; A module for obtaining the content of chemical components of tobacco leaves of a preset type, which is used to obtain near-infrared spectra of the samples by using a spectral acquisition instrument to acquire spectra of the sample set, and to obtain the content of chemical components of tobacco leaves of a preset type of each sample in the sample set by using a prediction model of a near-infrared big data platform based on the near-infrared spectra; A correlation analysis module, for performing grey correlation analysis on a preset number of sensory indicators and a preset type of tobacco chemical components, and finding chemical indicators with a correlation degree of a first preset percentage or above as a first indicator set; A model module is established, and a preset number of sensory indicators are used as dependent variables and the first indicator set is used as an independent variable. Prediction models of the sensory indicators are established respectively using the principal component regression analysis method. The principal component regression analysis method specifically includes: performing principal component regression according to the variance contribution percentage in the first indicator set, and when the cumulative variance contribution percentage reaches a second preset percentage or above, determining the final number of principal components related to the sensory indicators.

9. An electronic device, comprising: processor; as well as, A memory arranged to store computer executable instructions, which, when executed, cause the processor to perform the steps of the method for constructing a tobacco leaf raw material sensory quality prediction model as described in any one of claims 1 to 7.

10. A storage medium for storing computer executable instructions, which, when executed, implement the steps of the method for constructing a tobacco leaf raw material sensory quality prediction model as described in any one of claims 1 to 7.

Images