A general modeling method for conventional chemical components of tobacco leaves based on near-infrared spectroscopy

Abstract

Provided is a method for modeling a general model of conventional chemical components of tobacco leaves based on near-infrared spectroscopy, belonging to the technical field of cigarette manufacturing. The invention selects cut tobacco and tobacco dust as research objects, establishes a technical system for rapid quality analysis and evaluation of cut tobacco and tobacco dust by using near-infrared molecular vibration spectroscopy technology, combines pretreatment and variable selection to establish quantitative models of total alkaloids, total nitrogen, total sugar, reducing sugar, potassium ion and chloride ion of cut tobacco and tobacco dust, realizes rapid determination of quality indicators of cut tobacco and tobacco dust, provides a new method for rapid analysis and evaluation of quality indicators of cut tobacco and tobacco dust, and provides strong technical support for rapid, accurate and digital quality control of cut tobacco and tobacco dust in cigarette enterprises.

Description

Technical Field

[0001] The invention belongs to the technical field of cigarette manufacturing, and in particular relates to a method for modeling a general model of conventional chemical components of tobacco leaves based on near infrared spectroscopy. Background Art

[0002] As a consumer product that is becoming increasingly popular, the quality of cigarettes will directly affect the vital interests of consumers. Tobacco leaves and cut tobacco raw materials are the basis of cigarette production, and their intrinsic quality is the standard for measuring the availability of tobacco leaves and cut tobacco. The chemical composition of cigarettes has a great correlation with their intrinsic quality. Therefore, the detection of the main chemical components of cut tobacco leaves and cut tobacco in the cigarette making process has important theoretical significance and practical value for quality control of the cigarette industry.

[0003] According to the industry standard method, the continuous flow method is used to determine the content of total sugar, reducing sugar, chlorine, nicotine, potassium and total nitrogen. Although the continuous flow method can better control the reaction and have better reaction reproducibility, it has too many problems for batch reactions and cannot be handled on a larger scale. In addition, the analysis technology has disadvantages such as cumbersome pre-treatment steps, long analysis time, slow analysis speed, large consumption of manpower and material resources, and the methods are all selective. Data analysis requires more professional technicians, especially since these components cannot be measured simultaneously, which cannot meet the needs of rapid quality control of tobacco leaves.

[0004] At present, chromatography technology is gradually being applied to the detection of water-soluble sugars in tobacco leaves. In existing chromatography and mass spectrometry studies, although the detection limit and sensitivity of the methods are good, each method has its limitations: (1) it cannot simultaneously determine the total alkali content, total sugar content, reducing sugar content, total nitrogen content, potassium and chlorine in tobacco leaves; (2) the samples require complex pretreatment, which is time-consuming and labor-intensive, and there are still problems such as long analysis time, inability to perform rapid analysis, and large method limitations.

[0005] Near-infrared molecular vibration spectroscopy is mainly based on the absorption of the harmonic and summed frequencies of the XH vibration of hydrogen-containing groups. It has the advantages of being simple, fast (up to 3 minutes / sample), pollution-free, and low-cost. The spectrum contains rich structural information and strong fingerprint characteristics, which is suitable for the identification and identification of complex system samples. Near-infrared analysis technology has been widely used in the tobacco industry, such as origin tracing, tobacco leaf quality determination, and rapid prediction of the values ​​of conventional physical and chemical indicators of tobacco leaf quality (total nitrogen, nicotine, total sugar, etc.). Although near-infrared has been used in the determination of total alkaloids, total nitrogen, total sugar, reducing sugar, potassium ions, and chloride ions in tobacco cuts and tobacco ends, in these studies, the variable selection is only the simplest spectral segment splitting, which lacks theoretical and statistical significance. Therefore, it is necessary to propose improvements. Summary of the invention

[0006] The technical problem solved by the present invention is to provide a general modeling method for the conventional chemical components of tobacco leaves based on near-infrared spectroscopy. The present invention is proposed to address the problem that in the prior art, the variable selection in the modeling research on the conventional chemical components of tobacco leaves based on near-infrared molecular technology is only the simplest spectral band splitting, which lacks theoretical and statistical significance.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] The method for modeling a general model of conventional chemical components of tobacco leaves based on near infrared spectroscopy includes the following steps:

[0009] Step 1): Select different tobacco leaves, prepare tobacco leaf samples and store them for later use;

[0010] Step 2): collecting near infrared spectral data of each tobacco leaf sample and optimizing it into a near infrared spectral spectrum;

[0011] Step 3): Detecting the content of conventional chemical components in the tobacco leaf samples;

[0012] Step 4): set a set of prediction set target evaluation parameters for conventional chemical components of tobacco leaves; based on the near-infrared spectrum data of the samples collected in step 1), use a suitable preprocessing method to process the spectrum data; then perform a variable feature selection algorithm optimization model on the preprocessed spectrum data and the conventional chemical component content data of the tobacco leaf samples in step 2), complete the extraction of characteristic wavelengths of the conventional chemical components of the tobacco leaf samples, and compare them with the prediction set target evaluation parameters to determine the optimal feasible general model;

[0013] There are two types of variable feature selection algorithms: one is the wavelength interval optimization method, including interval partial least squares iPLS and moving window partial least squares MWPLS; the other is the wavelength point variable selection optimization method, including continuous projection algorithm SPA, irrelevant information variable elimination UVE, tandem algorithm UVE-SPA and wavelength interval combination optimization algorithm ICO;

[0014] Step 5): For the tobacco leaf samples to be tested, first collect their near-infrared spectral data and use the general model to predict the conventional chemical composition.

[0015] In the above step 1), when the tobacco leaf samples are prepared, the selected tobacco leaves are ground, labeled, and packed into sealed bags to balance the moisture content and store them for testing.

[0016] In the above step 2), when collecting near-infrared spectral data for each tobacco leaf sample, the sample is placed in a rotating sampling cup, and each sample is resampled three times, and the spectrum is recollected each time and saved for subsequent data processing.

[0017] In the above step) 2, the near-infrared spectrum measurement method adopts the diffuse reflectance integrating sphere sampling method; the near-infrared instrument working condition setting and spectrum acquisition: the Thermo Antaris II Fourier transform near-infrared spectrometer is used, and after preheating for 30 minutes, the spectrum scanning range is 10000-4000cm -1 , select resolution 8cm -1 , the scanning number is 64 times, the experimental environment temperature is (23±1)℃, the humidity is (50±2)%; the thickness of the sample held in the rotating sample cup is determined to be 0.5cm.

[0018] In the above step 3), the conventional chemical components are total plant alkaloids, total nitrogen, total sugars, reducing sugars, potassium ions and chloride ions.

[0019] In the above step 4), the specific process of each variable feature selection algorithm is:

[0020] (1) Interval Partial Least Squares (iPLS)

[0021] The interval partial least squares method is used to extract characteristic bands. It divides the entire spectral region into multiple sub-intervals of equal width, assuming that there are n of them. Partial least squares regression PLSR modeling is performed on each sub-interval to establish a "local regression model" of the quality to be tested, that is, a PLS model is established in each interval to obtain n local regression models. The root mean square error (MSE) value is used as the accuracy measurement standard for each model, and the sub-interval where the local model with the highest accuracy is located is taken as the characteristic band. These selected characteristic bands are merged to re-establish an optimized PLSR model.

[0022] (2) Moving Window Partial Least Squares (MWPLS)

[0023] The moving window partial least squares method is to continuously move a window of fixed width n along the spectral axis, moving one wavelength point each time, and establish a PLS model by cross-validation. In the process of moving, a series of PLS ​​models of different moving wavelength point windows are obtained. The residual square error PRESS or SSR corresponding to the number of main factors is used to evaluate the quality of the model, determine the variable interval with good effect of the moving window PLS model, select the spectral interval with high information content related to the component to be measured, and then merge all the selected variable intervals to re-establish an optimized PLS model.

[0024] (3) Continuous Projection Algorithm SPA

[0025] The SPA algorithm starts from a certain wavelength point and continuously adopts a continuous projection strategy to select the wavelength point with the smallest linear correlation with the existing wavelength to form a wavelength subset. Repeat the above operation to select a series of wavelength subsets, and then select the optimal wavelength subset by comparing the prediction ability of the model after modeling these wavelength subsets. This algorithm can reduce the collinearity problem between the selected wavelengths as much as possible.

[0026] (4) Irrelevant Information Variable Elimination UVE

[0027] Irrelevant information variable elimination UVE combines the spectrum matrix with the random noise signal matrix of equal length and then performs PLS regression, calculates the regression coefficient of each wavelength point to the PLS model, and removes the threshold variable whose contribution to the regression coefficient is lower than the maximum contribution of all random noise to the model, thereby achieving variable screening;

[0028] The threshold c is defined as:

[0029]

[0030] m is the number of points in each spectral data, and also the number of columns in the spectral matrix.

[0031] and s(i) are the mean vector and standard deviation of the regression coefficient matrix b, respectively, taking the maximum absolute value of c within the random noise variable interval:

[0032] C max =max[abs(c)]

[0033] Eliminate the variables with c≤Cmax in the spectral matrix, and perform subsequent operations on the new matrix XUVE composed of the remaining variables;

[0034] (5) Series algorithm UVE-SPA: used to organically integrate the irrelevant information variable elimination UVE algorithm and the continuous projection algorithm SPA;

[0035] (6) Wavelength Interval Combination Optimization Algorithm ICO

[0036] The wavelength interval combination optimization algorithm ICO breaks the boundary between wavelength point selection and wavelength interval selection. As an interval optimization selection algorithm developed under the MPA framework, the calculation process of the ICO algorithm is summarized as follows:

[0037] 1) Divide the near-infrared spectrum of the sample into N equal parts of roughly equal width, each part being a spectral interval;

[0038] 2) Using WBS to generate a subset (e.g., 1000) of M random combinations of different wavelength intervals. It should be noted that the initial sampling weight of each wavelength point is 1 by default;

[0039] 3) The PLS algorithm and 5-fold interaction test were used to calculate the RMSECV value corresponding to each interval combination subset;

[0040] 4) Extract a certain proportion of the optimal interval combination subset from all interval combinations, and calculate the average value of the RMSECV value corresponding to this part of the interval combination subset, recorded as RMSECV m ; The above ratio is denoted as α, which can usually be set to 0.1, or it can be optimized according to the actual situation of the data.

[0041] The advantages of the present invention compared with the prior art are:

[0042] 1. This scheme selects tobacco leaves as the research object, uses near-infrared molecular vibration spectroscopy technology to establish a technical system for rapid quality analysis and evaluation of tobacco leaves, and combines pretreatment and variable selection to establish quantitative models for total alkaloids, total nitrogen, total sugar, reducing sugar, potassium ions, and chloride ions in tobacco cuts and tobacco dust, so as to achieve rapid determination of tobacco leaf quality indicators, provide a new method for rapid analysis and evaluation of tobacco leaf quality indicators, and provide strong technical support for the rapid, accurate, and digital quality control of tobacco leaves in cigarette enterprises;

[0043] 2. This scheme uses interval partial least squares iPLS, moving window partial least squares MWPLS, continuous projection algorithm SPA, irrelevant information variable elimination UVE, tandem algorithm UVE-SPA and wavelength interval combination optimization algorithm ICO feature selection algorithm to complete the PLS model of total sugar, reducing sugar, total nitrogen, total alkali, potassium and chlorine in tobacco leaves, extract relevant characteristic wavelengths, and thus establish a better and more chemically meaningful model. By optimizing the model through wavelength selection, the model performance is improved, the model is simplified, the amount of calculation is reduced, the model's interpretability is improved, and the model's robustness is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The near infrared spectrum of the tobacco leaf sample in the present invention Figure 1 ;

[0045] Figure 2 The near infrared spectrum of the tobacco leaf sample in the present invention Figure 2 ;

[0046] Figure 3 The near infrared spectrum of the tobacco leaf sample in the present invention Figure 3 ;

[0047] Figure 4 The near infrared spectrum of the tobacco leaf sample in the present invention Figure 4 ;

[0048] Figure 5 The near infrared spectrum of the tobacco leaf sample in the present invention Figure 5 ;

[0049] Figure 6 The near infrared spectrum of the tobacco leaf sample in the present invention Figure 6 ;

[0050] Figure 7 It is a flow chart of the wavelength interval combination optimization algorithm ICO in the present invention. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] See also Figure 1-7 , details of the embodiments of the present invention.

[0053] Example 1: A method for modeling a general model of conventional chemical components of tobacco leaves based on near infrared spectroscopy, specifically comprising the following steps:

[0054] Step 1): more than 300 tobacco leaf samples in stock from a cigarette production area are selected, ground, labeled, and packed into sealed bags to balance the moisture content and store for testing.

[0055] Step 2): Collect near infrared spectral data of each tobacco leaf sample and optimize it into a near infrared spectral spectrum. Figure 1-6 As shown;

[0056] The near-infrared spectrum measurement method adopts the diffuse reflectance integrating sphere sampling method; the near-infrared instrument working condition setting and spectrum acquisition: the Thermo Antaris II Fourier transform near-infrared spectrometer is used, and after preheating for 30 minutes, the spectrum scanning range is 10000-4000cm -1 , select resolution 8cm -1 , scanning times 64 times, experimental environment temperature is (23±1)℃, humidity is (50±2)%;

[0057] Since near-infrared light has a relatively strong penetrating ability, the experiment examined the effect of the thickness of the sample in the rotating sample cup on the scanning results. The smaller the difference in the absorbance values ​​of samples with similar thickness settings, the smaller the difference between the spectra. It was finally determined that when the thickness was about 0.5 cm, the spectral quality was stable.

[0058] When collecting near-infrared spectral data for each tobacco leaf sample, the sample was placed in a rotating sampling cup, and each sample was resampled three times, with the spectrum recollected each time. A total of about 1,200 near-infrared spectra were collected and saved for subsequent data processing.

[0059] Step 3): Use the flow method to detect the content of conventional chemical components in the tobacco leaf sample; specifically, the conventional chemical components are total alkaloids, total nitrogen, total sugars, reducing sugars, potassium ions, chloride ions, etc.

[0060] Step 4): set a set of prediction set target evaluation parameters for conventional chemical components of tobacco leaves; based on the near-infrared spectrum data of the samples collected in step 1), use a suitable preprocessing method to process the spectrum data; then perform a variable feature selection algorithm optimization model on the preprocessed spectrum data and the conventional chemical component content data of the tobacco leaf samples in step 2), complete the extraction of characteristic wavelengths of the conventional chemical components of the tobacco leaf samples, and compare them with the prediction set target evaluation parameters to determine the optimal feasible general model;

[0061] Among them, the variable feature selection algorithms include two categories. One is the wavelength interval optimization method, including interval partial least squares iPLS and moving window partial least squares MWPLS; the other is the wavelength point variable selection optimization method, including continuous projection algorithm SPA, irrelevant information variable elimination UVE, tandem algorithm UVE-SPA and wavelength interval combination optimization algorithm ICO.

[0062] The present invention adopts interval partial least squares iPLS, moving window partial least squares MWPLS, continuous projection algorithm SPA, irrelevant information variable elimination UVE, tandem algorithm UVE-SPA and wavelength interval combination optimization algorithm ICO feature selection algorithm to complete the PLS model of total sugar, reducing sugar, total nitrogen, total alkali, potassium and chlorine in tobacco and tobacco dust, extract relevant characteristic wavelengths, and thus establish a better and more chemically meaningful model. By performing wavelength selection optimization model, the model performance is improved, the model is simplified, the amount of calculation is reduced, the interpretability of the model is improved, and the robustness of the model is improved.

[0063] Specifically, the specific process of each variable feature selection algorithm is:

[0064] (1) Interval Partial Least Squares (iPLS)

[0065] The interval partial least squares method is used to extract characteristic bands. It divides the entire spectral region into multiple sub-intervals of equal width, assuming that there are n of them. Partial least squares regression PLSR modeling is performed on each sub-interval to establish a "local regression model" of the quality to be tested, that is, a PLS model is established in each interval to obtain n local regression models. The root mean square error (MSE) value is used as the accuracy measurement standard for each model, and the sub-interval where the local model with the highest accuracy is located is taken as the characteristic band. These selected characteristic bands are merged to re-establish an optimized PLSR model.

[0066] The indicators formed by iPLS modeling with non-interval partial least squares are shown in the following table:

[0067] iPLS nlv Mean RMSEC <![CDATA[CalQ 2 ]]> RMSEP <![CDATA[PreQ 2 ]]> Potassium 10 2.3548 0.3022 0.6130 0.2891 0.6409 nitrogen 10 2.1590 0.1342 0.8038 0.1027 0.8794 Total Sugar 9 25.6625 2.3485 0.8139 2.0880 0.8475 Reducing sugar 10 20.9808 0.2518 0.8765 0.2370 0.8986 Chloride ion 10 0.4163 0.2271 0.9023 0.2143 0.9186 nicotine 10 2.3156 0.2274 0.9028 0.2817 0.8466

[0068] As can be seen from the table above, using the iPLS model, reducing sugar and nicotine have the best effect, the correlation coefficient of the calibration set is higher than the correlation coefficient of the prediction set, and the standard deviation is lower. The correlation coefficient of potassium ion is the lowest. The standard deviation of chloride ion is larger.

[0069] (2) Moving Window Partial Least Squares (MWPLS)

[0070] The moving window partial least squares method is to continuously move a window of fixed width n along the spectral axis, moving one wavelength point each time, and establish a PLS model using cross-validation. During the moving process, a series of PLS ​​models with different moving wavelength point windows are obtained. The residual square error PRESS or SSR corresponding to the number of main factors is used to evaluate the model, determine the variable interval with good effect of the moving window PLS model, select the spectral interval with high information content related to the component to be measured, and then merge all the selected variable intervals to rebuild an optimized PLS model.

[0071] The indicators formed by the moving window partial least squares MWPLS modeling are shown in the following table:

[0072] MWPLS nlv Mean RMSEC <![CDATA[CalQ 2 ]]> RMSEP <![CDATA[PreQ 2 ]]> Potassium 10 2.3548 0.2463 0.8851 0.2181 0.9134 nitrogen 10 2.1590 0.2485 0.8813 0.2788 0.8591 nicotine 10 25.6625 0.2256 0.9044 0.2896 0.8378 Reducing sugar 10 20.9808 0.2158 0.9094 0.2102 0.9204 Chloride ion 10 0.4163 0.2213 0.9072 0.2027 0.9272 Total Sugar 10 2.3156 0.2193 0.9111 0.2813 0.8432

[0073] As can be seen from the table above, the MWPLS model has the best modeling effect for reducing sugars, with the calibration set determination coefficient and the prediction set determination coefficient both above 0.9. Although the calibration set and prediction set determination coefficients for chloride ions are high, the standard deviation is high.

[0074] (3) Continuous Projection Algorithm SPA

[0075] The SPA algorithm starts from a certain wavelength point and continuously uses a continuous projection strategy to select the wavelength point with the smallest linear correlation with the existing wavelength to form a wavelength subset. The above operation is repeated to select a series of wavelength subsets, and then the optimal wavelength subset is selected by comparing the prediction ability of the model after modeling these wavelength subsets. This algorithm can reduce the collinearity problem between the selected wavelengths as much as possible.

[0076] (4) Irrelevant Information Variable Elimination UVE

[0077] Irrelevant information variable elimination UVE combines the spectrum matrix with the random noise signal matrix of equal length and then performs PLS regression, calculates the regression coefficient of each wavelength point to the PLS model, and removes the threshold variable whose contribution to the regression coefficient is lower than the maximum contribution of all random noise to the model, thereby achieving variable screening;

[0078] The threshold c is defined as:

[0079]

[0080] m is the number of points in each spectral data, and also the number of columns in the spectral matrix.

[0081] and s(i) are the mean vector and standard deviation of the regression coefficient matrix b, respectively, taking the maximum absolute value of c within the random noise variable interval:

[0082] C max =max[abs(c)]

[0083] Eliminate the variables with c≤Cmax in the spectral matrix, and perform subsequent operations on the new matrix XUVE composed of the remaining variables.

[0084] The indicators formed after UVE modeling by eliminating irrelevant information variables are shown in the following table.

[0085] UVE Mean nlv RMSEC <![CDATA[CalQ 2 ]]> RMSEP <![CDATA[PreQ 2 ]]> Potassium 2.3548 10 0.2938 0.6731 0.2119 -0.5435 nitrogen 2.1590 10 0.1368 0.8088 0.0703 0.7434 Total Sugar 25.6625 10 2.5607 0.7720 2.0799 0.3980 Reducing sugar 20.9808 10 1.7725 0.8184 1.7967 0.8103 Chloride ion 0.4163 10 0.0810 0.6692 0.0834 -0.4670 nicotine 2.3156 10 0.2680 0.8654 0.2645 0.3521

[0086] As can be seen from the table above, using the UVE model, the standard deviations of the indicators are not much different. The modeling effect of potassium ion and chloride ion is poor, the determination coefficient of the calibration set is small, and the determination coefficient of the prediction set is even negative. The modeling effect of reducing sugar is good, and the determination coefficient of the calibration set and the determination coefficient of the prediction set are both above 0.8.

[0087] (5) Series Algorithm UVE-SPA: It is used to organically integrate the UVE algorithm for eliminating irrelevant information variables and the continuous projection algorithm SPA. In terms of the process, UVE is performed first, and then SPA is performed. The two algorithms select important variables from different angles. UVE selects useful variables, but it does not care whether these variables are related; the SPA algorithm specifically selects irrelevant variables, but it does not consider whether these variables are important.

[0088] (6) Wavelength Interval Combination Optimization Algorithm ICO

[0089] The wavelength interval combination optimization algorithm ICO breaks the boundary between wavelength point selection and wavelength interval selection. As an interval optimization selection algorithm developed under the MPA framework, the calculation process of the ICO algorithm is summarized as follows:

[0090] 1) Divide the near-infrared spectrum of the sample into N equal parts of roughly equal width, each part being a spectral interval;

[0091] 2) Using WBS to generate a subset (e.g., 1000) of M random combinations of different wavelength intervals. It should be noted that the initial sampling weight of each wavelength point is 1 by default;

[0092] 3) The PLS algorithm and 5-fold interaction test were used to calculate the RMSECV value corresponding to each interval combination subset;

[0093] 4) Extract a certain proportion of the optimal interval combination subset from all interval combinations, and calculate the average value of the RMSECV value corresponding to this part of the interval combination subset, recorded as RMSECV m ; The above ratio is recorded as α, which can usually be set to 0.1, or it can be optimized according to the actual situation of the data. Figure 7 Shown is the flow chart of the wavelength interval combinatorial optimization algorithm ICO, in which Spectra splitting: spectral partitioning, WBS sampling weights: weighted bootstrap sampling weights, Sub-models generation: sub-model generation, Model extraction: model optimization, New weights: new weights, Rmsecv: verification root mean square standard deviation, Local Search: local search, Finally selectedwavelength: finally selected wavelength.

[0094] From the above models, it can be seen that the modeling effect of reducing sugar is the best, with a large determination coefficient and a small standard deviation. The modeling effect of potassium ion is the worst, with a small determination coefficient. Although the determination coefficient of chloride ion is acceptable, the standard deviation is high.

[0095] From the above, it can be seen that MWPLS has the best modeling effect, the determination coefficients of the six chemical values ​​are closest to 1, and the standard deviation is the lowest, so the MWPLS algorithm is selected.

[0096] Step 5): For the tobacco leaf samples to be tested, first collect their near-infrared spectral data and use the general model to predict the conventional chemical composition.

[0097] In summary, the present invention selects tobacco leaves as the research object, uses near-infrared molecular vibration spectroscopy technology to establish a technical system for rapid quality analysis and evaluation methods of tobacco leaves, combines pretreatment and variable selection to establish quantitative models of total alkaloids, total nitrogen, total sugar, reducing sugar, potassium ions, and chloride ions of tobacco leaves, and realizes rapid determination of quality indicators of flue-cured tobacco leaves, provides a new method for rapid analysis and evaluation of quality indicators of tobacco shreds and tobacco dust, and provides strong technical support for the rapid, accurate, and digital quality control of tobacco shreds and tobacco dust in cigarette enterprises.

[0098] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0099] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A general modeling method for conventional chemical components of tobacco leaves based on near infrared spectroscopy, characterized in that: The following steps are involved: Step 1): Select different tobacco leaves, make tobacco leaf samples and store them for later use; Step 2): Collect near infrared spectral data of each tobacco leaf sample and optimize it into a near infrared spectral spectrum; Step 3): Detect the content of conventional chemical components in tobacco leaf samples; Step 4): Set a set of prediction set target evaluation parameters for conventional chemical components of tobacco leaves; Based on the near-infrared spectral data of the samples collected in step 1), use a suitable preprocessing method to process the spectral data; Then, perform a variable feature selection algorithm optimization model on the preprocessed spectral data and the conventional chemical component content data of the tobacco leaf samples in step 2), complete the extraction of characteristic wavelengths of the conventional chemical components of the tobacco leaf samples, and compare them with the prediction set target evaluation parameters to determine the optimal feasible general model; There are two types of variable feature selection algorithms: one is the wavelength interval optimization method, including interval partial least squares iPLS and moving window partial least squares MWPLS; the other is the wavelength point variable selection optimization method, including continuous projection algorithm SPA, irrelevant information variable elimination UVE, tandem algorithm UVE-SPA and wavelength interval combination optimization algorithm ICO; Step 5): For the tobacco leaf samples to be tested, first collect their near-infrared spectrum data and use the general model to predict the conventional chemical composition; In the above step 2), the near-infrared spectrum measurement method adopts the diffuse reflectance integrating sphere sampling method; the near-infrared instrument working condition setting and spectrum acquisition: the Thermo Antaris II Fourier transform near-infrared spectrometer is used, and after preheating for 30 minutes, the spectrum scanning range is 10000-4000cm -1 , select resolution 8cm -1 , the number of scans was 64 times, the experimental environment temperature was (23±1)℃, the humidity was (50±2)%; the thickness of the sample held in the rotating sample cup was determined to be 0.5cm; In the above step 3), the conventional chemical components are total plant alkaloids, total nitrogen, total sugars, reducing sugars, potassium ions and chloride ions; In the above step 4), the specific process of each variable feature selection algorithm is: (1) Interval Partial Least Squares (iPLS) The interval partial least squares method is used to extract characteristic bands. It divides the entire spectral region into multiple sub-intervals of equal width, assuming that there are n of them. Partial least squares regression PLSR modeling is performed on each sub-interval to establish a "local regression model" of the quality to be tested, that is, a PLS model is established in each interval to obtain n local regression models. The root mean square error (MSE) value is used as the accuracy measurement standard for each model, and the sub-interval where the local model with the highest accuracy is located is taken as the characteristic band. The selected characteristic bands are merged to re-establish an optimized PLSR model. (2) Moving Window Partial Least Squares (MWPLS) The moving window partial least squares method is to continuously move a window of fixed width n along the spectral axis, moving one wavelength point each time, and establish a PLS model by cross-validation. In the process of moving, a series of PLS ​​models of different moving wavelength point windows are obtained. The residual square error PRESS or SSR corresponding to the number of main factors is used to evaluate the quality of the model, determine the variable interval with good effect of the moving window PLS model, select the spectral interval with high information content related to the component to be measured, and then merge all the selected variable intervals to re-establish an optimized PLS model. (3) Continuous Projection Algorithm SPA The SPA algorithm starts from a certain wavelength point and continuously adopts a continuous projection strategy to select the wavelength point with the smallest linear correlation with the existing wavelength to form a wavelength subset. It repeats and selects a series of wavelength subsets, and then selects the optimal wavelength subset by comparing the prediction ability of the model after modeling these wavelength subsets. (4) Elimination of irrelevant information variables UVE Irrelevant information variable elimination UVE combines the spectrum matrix with the random noise signal matrix of equal length and then performs PLS regression, calculates the regression coefficient of each wavelength point to the PLS model, and removes the threshold variable whose contribution to the regression coefficient is lower than the maximum contribution of all random noise to the model, thereby achieving variable screening; The threshold c is defined as: m is the number of points in each spectral data, and also the number of columns in the spectral matrix. and s(i) are the mean vector and standard deviation of the regression coefficient matrix b, respectively, taking the maximum absolute value of c within the random noise variable interval: Eliminate the variables with c≤Cmax in the spectral matrix, and perform subsequent operations on the new matrix XUVE composed of the remaining variables; (5) Series algorithm UVE-SPA: used to organically integrate the UVE algorithm for eliminating irrelevant information variables and the continuous projection algorithm SPA; (6) Wavelength interval combination optimization algorithm ICO The wavelength interval combination optimization algorithm ICO breaks the boundary between wavelength point selection and wavelength interval selection. As an interval optimization selection algorithm developed under the MPA framework, the calculation process of the ICO algorithm is summarized as follows: 1) Divide the near-infrared spectrum of the sample into N equal parts of roughly equal width, each of which is a spectral interval; 2) Use WBS to generate a subset of M random combinations of different wavelength intervals. It should be noted that the initial sampling weight of each wavelength point is 1 by default; 3) The PLS algorithm and 5-fold interaction test were used to calculate the RMSECV value corresponding to each interval combination subset; 4) Extract a certain proportion of the optimal interval combination subset from all interval combinations, and calculate the average value of the RMSECV value corresponding to this part of the interval combination subset, recorded as RMSECV m ; Among them, the above ratio is denoted as α, and α is set to 0.

1.

2. The method for modeling a general model of conventional chemical components of tobacco leaves based on near infrared spectroscopy according to claim 1, characterized in that: In the above step 1), when preparing tobacco leaf samples, different tobacco leaves are selected, ground, labeled, and packed into sealed bags to balance the moisture content and store for testing.

3. The method for modeling a general model of conventional chemical components of tobacco leaves based on near infrared spectroscopy according to claim 1, characterized in that: In the above step 2), when collecting near-infrared spectral data for each tobacco leaf sample, the sample is placed in a rotating sampling cup, and each sample is resampled three times, and the spectrum is recollected each time and saved for subsequent data processing.

Images