Optimization method of chemical composition model of tobacco leaf near infrared spectrum based on transfer learning

Abstract

The present invention relates to a method for optimizing a tobacco leaf near-infrared spectroscopy chemical component model based on transfer learning, and belongs to the technical field of tobacco leaf detection. The present invention provides a method for optimizing a tobacco leaf near-infrared spectroscopy chemical component model based on transfer learning. On the basis of an existing near-infrared spectroscopy tobacco detection model, in order to cope with the accuracy impact of the original model due to changes in factors such as environment and climate, the method of correlation-ratio transfer learning is used to transfer effective information in the broken leaf spectrum, improve the stability of the whole leaf spectrum measurement model, and establish a new prediction model for total sugar, reducing sugar, total alkaloids, potassium, and chlorine content. The optimization method described in the present invention can reduce the workload consumed by resampling modeling, establish a more constant prediction model, can be used for the analysis of tobacco leaf chemical components in the tobacco production process, meet the requirements for rapid detection of conventional chemical components of tobacco leaves, and the data obtained can meet the needs of tobacco production enterprises.

Description

Technical Field

[0001] The present invention belongs to the technical field of tobacco leaf component detection, and in particular relates to a method for optimizing a tobacco leaf near-infrared spectroscopy chemical component model based on transfer learning. Background Art

[0002] Near infrared spectroscopy (NIR) refers to electromagnetic waves between visible light (VIS) and mid-infrared (MIR) spectral regions, with a wavelength between 780 and 2526 nm, which can be divided into short-wave (780-1100 nm) and long-wave (1100-2526 nm) regions. Near infrared spectral information can reflect the double frequency absorption and combined frequency absorption of hydrogen-containing groups such as CH, OH, and NH in organic molecules. Different chemical components have different absorption of near infrared light, and the obtained spectra are also different. By directly or indirectly measuring the near infrared spectrum of the sample, according to their absorption conditions, the relationship between the content of a certain chemical component of the sample and its near infrared absorption spectrum, that is, the mathematical model between the two, can be established. Based on the established mathematical model, the near infrared spectral absorption data is used to predict the content of related chemical components of similar samples. This method only needs to obtain the near infrared spectral data of the sample, with fast detection speed, high timeliness, and no damage to the sample. At present, NIR analysis technology has been applied to many fields such as agriculture, petrochemicals, pharmaceuticals, food, and clinical.

[0003] Tobacco contains many chemical components, among which total sugar, reducing sugar, total nitrogen, total alkaloids, potassium, chlorine and other indicators are the most basic chemical component indicators for evaluating tobacco leaf quality, and play a very important role in the acquisition and grading of tobacco leaves, as well as the formulation design and quality control of cigarettes. The detection of chemical components in tobacco by near-infrared spectroscopy has gradually become popular and promoted in tobacco production. Many researchers have conducted research on this, and have formed a good mathematical model and achieved many research results, which has promoted the development of the tobacco industry.

[0004] Chinese patent number CN114088661A discloses an online prediction method for the chemical composition of tobacco leaves in the baking process based on transfer learning and near-infrared spectroscopy. The method collects tobacco leaf samples multiple times and iteratively trains a tobacco leaf chemical composition prediction model through algorithm analysis data. The chemical composition prediction model is used to obtain the prediction results of the tobacco leaf chemical composition.

[0005] Since the chemical composition of whole tobacco leaves is unevenly distributed, and portable spectrometers often use point-to-surface methods to collect spectral information, the model stability is poor. After the whole tobacco leaves are processed into broken leaves and flattened, the chemical composition becomes more "uniform". Therefore, the above technical problems can be solved by obtaining the near-infrared spectra of the broken leaves and using the correlation ratio transfer learning algorithm to establish a whole-leaf tobacco component analysis model. Summary of the invention

[0006] In order to overcome the defect that a portable spectrometer cannot comprehensively collect spectral information of a whole tobacco leaf, the present invention provides a method for optimizing a tobacco leaf near-infrared spectral chemical composition model based on transfer learning. The method takes into account the effective information of migrating broken leaf spectra, applies the transfer learning method to the existing component prediction model, and improves the accuracy of the whole leaf spectral measurement model, making it more suitable for rapid detection requirements of conventional chemical components of tobacco leaves. The main purpose is to establish a tobacco leaf chemical composition prediction model that can directly perform model migration and is more stable, thereby reducing the workload of re-collecting new samples and establishing models, and achieving relatively high prediction accuracy when using the spectral information of whole leaf samples, thereby providing a more constant prediction model for conventional chemical component detection of tobacco leaves.

[0007] To achieve the above objectives, the present invention provides a method for optimizing a tobacco leaf near-infrared spectral chemical composition model. A tobacco leaf chemical composition analysis model is established based on near-infrared spectral data of whole and broken tobacco leaves. The effective information of the broken tobacco leaf spectra is transferred to the whole leaf spectral model. A tobacco leaf near-infrared spectral chemical composition optimization model based on transfer learning is established. The above optimization model is then used to perform component detection on the tobacco leaves to be tested.

[0008] The technical solution adopted by the present invention is: a method for optimizing a tobacco leaf near-infrared spectroscopy chemical composition model based on transfer learning, the optimization method comprising the following steps:

[0009] Step S1, obtaining a near infrared spectrum of a whole tobacco leaf;

[0010] Step S2, obtaining near infrared spectrum of broken tobacco leaves;

[0011] Step S3, obtaining reference chemical composition values ​​by drying and grinding the tobacco leaves and testing them with a flow analyzer;

[0012] Step S4, preprocessing the near infrared spectrum data obtained in step S1 and step S2;

[0013] Step S5, establishing chemical composition analysis models for whole tobacco leaves and broken tobacco leaves respectively according to the near infrared spectrum data of step S1 and step S2;

[0014] Step S6, transferring the broken leaf spectral information to the whole leaf spectral model through the association-ratio transfer learning method, so as to obtain the optimized model of the chemical composition of the tobacco leaf near-infrared spectrum; detecting the chemical composition of the tobacco leaf according to the optimized model, and obtaining the predicted value of the chemical composition of the tobacco leaf.

[0015] Furthermore, in step S6, the optimization model of the tobacco leaf near-infrared spectrum chemical components is implemented by the following steps:

[0016] Step S61: firstly, a portable tobacco chemical component analyzer is used to scan a tobacco leaf sample to collect spectrum data information of the whole leaf;

[0017] Step S62, using a portable tobacco chemical component analyzer to scan the crushed tobacco leaf samples and collect spectral data information of the crushed leaves;

[0018] Step S63, using a continuous flow method to determine the chemical component value information of the total sugar, reducing sugar, total alkaloids, potassium, and chlorine content of the first-cured tobacco leaf sample;

[0019] Step S64, pre-processing the spectral data using smoothing, derivative method, and standard normalization (SNV) processing methods in sequence to reduce the influence of noise and system errors on the spectrum;

[0020] Step S65, select partial least squares (PLS) to build a model. The sum of squares of the prediction residuals of PLS ​​is small and has high prediction stability.

[0021] Step S66: Based on the model obtained by PLS modeling in step S65, the association-ratio transfer learning method is used to calculate the parameter β of the optimization model according to the corresponding formula, so as to obtain the required optimization prediction model.

[0022] Furthermore, the method for obtaining the predicted value of the tobacco leaf chemical composition is achieved by the following steps:

[0023] Step S67, collecting spectrum information of the whole tobacco leaf;

[0024] Step S68, preprocessing the sample spectrum data obtained in step S67;

[0025] Step S69: Substitute the data obtained in step S68 into the optimization model of the chemical components of tobacco leaves in near-infrared spectrum to obtain the predicted values ​​of the chemical components of tobacco leaves.

[0026] Furthermore, in step S66, the transfer learning modeling method is:

[0027] According to the model obtained by PLS modeling, the association-ratio transfer learning method is used. If there are models P and Q, where , ，

[0028] Response variable , ,

[0029] Covariates , ;

[0030] If models P and Q are both linear models, let:

[0031]

[0032] like ,and exist,

[0033] The model parameters have the following relationship:

[0034]

[0035] in , where Q is the laboratory data model, The spectral information obtained by scanning the whole leaf. The spectral information obtained from scanning the broken leaves, The tobacco chemical composition values ​​are determined by continuous flow method;

[0036] The model is obtained according to PLS modeling ,

[0037] Then according to the formula:

[0038] , find the parameters ;Computational model This is the prediction model we need. When predicting the chemical composition of tobacco leaves using near infrared spectroscopy, we only need to collect the spectral information of the whole tobacco leaf and bring it into the model P. The predicted values ​​of tobacco leaf chemical components can be obtained .

[0039] Furthermore, the method for obtaining the predicted value of the chemical components of the tobacco leaves also includes converting the predicted value of the chemical components of the tobacco leaves obtained in step S69 into an expression understandable to the user according to a conversion rule and outputting it.

[0040] Furthermore, the accuracy evaluation criteria for the prediction value obtained in step S69 include the internal cross-prediction determination coefficient R 2 Test and internal cross-root mean square RMSECV validation.

[0041] Furthermore, the preprocessing method for the spectral data in step S4 includes smoothing, derivative method, and standard normalization (SNV) processing method.

[0042] Furthermore, the reference chemical component values ​​include total sugar, reducing sugar, total alkaloids, total potassium and total chlorine values.

[0043] Furthermore, the near-infrared spectrometer used to obtain the infrared spectrum in step S1 and step S2 is a YKS-H1 portable tobacco chemical composition analyzer.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention mainly uses the partial least squares method (PLS) to build a model first, and then uses the correlation-ratio transfer learning method to optimize the model based on this model, so as to obtain a brand-new component prediction model. When predicting the chemical components of tobacco leaves by near-infrared spectroscopy, it is only necessary to collect the spectral information of the whole tobacco leaf to obtain the predicted value of the chemical components of the tobacco leaves. This method effectively solves the problem that the existing component detection model is not suitable for tobacco leaves of new years due to changes in environmental and climatic factors, reduces the workload of re-collecting samples and modeling; and solves the problem that the chemical components of the whole tobacco leaf are unevenly distributed, and the portable spectrometer often uses the point-to-surface method to collect spectral information, resulting in unsatisfactory model stability, and meets the operability conditions of whole leaf collection in actual production environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of the present invention;

[0047] Figure 2a This is the near infrared spectrum of the whole leaf;

[0048] Figure 2b This is the near-infrared spectrum of broken leaves. DETAILED DESCRIPTION

[0049] 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.

[0050] like Figure 1 to Figure 2b As shown, the present invention provides an embodiment of a method for optimizing a tobacco leaf near-infrared spectroscopy chemical composition model based on transfer learning.

[0051] The optimization method comprises the following steps:

[0052] Step S1, obtaining a near infrared spectrum of a whole tobacco leaf;

[0053] Step S2, obtaining near infrared spectrum of broken tobacco leaves;

[0054] Step S3, obtaining reference chemical composition values ​​by drying and grinding the tobacco leaves and testing them with a flow analyzer;

[0055] Step S4, preprocessing the near infrared spectrum data obtained in step S1 and step S2;

[0056] Step S5, establishing chemical composition analysis models for whole tobacco leaves and broken tobacco leaves respectively according to the near infrared spectrum data of step S1 and step S2;

[0057] Step S6, the broken leaf spectral information is transferred to the whole leaf spectral model through the association-ratio transfer learning method, so as to obtain the optimization model of the chemical components of the tobacco leaf near-infrared spectrum; the chemical components of the tobacco leaf are detected according to the optimization model, and the predicted values ​​of the chemical components of the tobacco leaf are obtained. The present invention provides a method for optimizing the chemical component model of tobacco leaf near-infrared spectrum based on transfer learning, which is based on the near-infrared spectral data of whole tobacco leaves and broken leaves, establishes a tobacco leaf chemical component analysis model, transfers the effective information of the broken tobacco leaf spectrum to the whole leaf spectral model, establishes a tobacco leaf near-infrared spectral chemical component optimization model based on transfer learning, and then uses the above optimization model to detect the components of the tobacco leaf to be tested. In the embodiment of the present invention, the near-infrared spectrometer for obtaining the infrared spectrum in the steps S1 and S2 is a YKS-H1 portable tobacco chemical component analyzer. The analyzer uses the world's leading MEMS micromirror technology, built-in high-performance long-life light source, working wavelength range 1350-2500nm, working temperature -5℃-40℃, can quickly and non-destructively detect tobacco leaves' total sugar, reducing sugar, nicotine, potassium, chlorine and other chemical indicators. The continuous flow analyzer used is the Dutch skalar SAN++.

[0058] In the embodiment of the present invention, in step S6, the optimization model of the chemical components of tobacco leaf near-infrared spectrum is implemented by the following steps:

[0059] Step S61: firstly, a portable tobacco chemical component analyzer is used to scan a tobacco leaf sample to collect spectrum data information of the whole leaf;

[0060] Step S62, using a portable tobacco chemical component analyzer to scan the crushed tobacco leaf samples and collect spectral data information of the crushed leaves;

[0061] Step S63, using a continuous flow method to determine the chemical component value information of the total sugar, reducing sugar, total alkaloids, potassium, and chlorine content of the first-cured tobacco leaf sample;

[0062] Step S64, pre-processing the spectral data using smoothing, derivative method, and standard normalization (SNV) processing methods in sequence to reduce the influence of noise and system errors on the spectrum;

[0063] Step S65, select partial least squares (PLS) to build a model. The sum of squares of the prediction residuals of PLS ​​is small and has high prediction stability.

[0064] Step S66: Based on the model obtained by PLS modeling in step S65, the association-ratio transfer learning method is used to calculate the parameter β of the optimization model according to the corresponding formula, so as to obtain the required optimization prediction model.

[0065] In the implementation of the present invention, the method for obtaining the predicted value of the tobacco leaf chemical composition is achieved by the following steps:

[0066] Step S67, collecting spectrum information of the whole tobacco leaf;

[0067] Step S68, preprocessing the sample spectrum data obtained in step S67;

[0068] Step S69: Substitute the data obtained in step S68 into the optimization model of the chemical components of tobacco leaves in near-infrared spectrum to obtain the predicted values ​​of the chemical components of tobacco leaves.

[0069] The method for obtaining the predicted value of the chemical components of the tobacco leaves also includes converting the predicted value of the chemical components of the tobacco leaves obtained in step S69 into an expression understandable to the user according to a conversion rule and outputting it.

[0070] In the embodiment of the present invention, (1) the tobacco leaf spectrum collected by the MEMS micromirror technology near-infrared spectrometer is used as the source domain spectrum, with a total of more than 300 spectra, and the measured wavelength range is 1350-2500nm;

[0071] (2) Determination of the chemical composition of tobacco leaf samples, including total sugar, reducing sugar, total alkaloids, potassium, and chlorine content. According to the national industry standards for the chemical values ​​of tobacco samples YC / T 159-2019, YC / T 468-2013, YC / T 217-2007, and YC / T 162-2011, a continuous flow analyzer was used to determine the total sugar and reducing sugar (YC / T 159-2019), total alkaloids (YC / T 468-2013), total potassium (YC / T 217-2007), and total chlorine (YC / T 162-2011) in tobacco leaf samples.

[0072] (3) Spectral preprocessing: First, smoothing is used, and then the derivative method and standard normalization (SNV) processing method are used to eliminate the interference caused by irrelevant factors, so as to reduce the influence of noise and systematic errors on the spectrum.

[0073] (4) Modeling method: First, partial least squares (PLS) was used to build the model. PLS combines three methods: multivariate linear regression, canonical correlation analysis, and principal component analysis. The obtained model has a small sum of squares of the prediction residuals and high prediction stability.

[0074] (5) Transfer method: In terms of using the effective information of broken leaf spectra, the transfer learning method has more prominent advantages than other technical methods. The established model is optimized using the association-ratio transfer learning method. The new prediction model established in this way can predict the chemical composition of tobacco leaves by near-infrared spectroscopy only through the spectral information of the whole tobacco leaves.

[0075] like Figure 1 As shown, the establishment of a tobacco leaf near-infrared spectroscopy chemical composition optimization model based on transfer learning is achieved through the following steps:

[0076] 1) First, a portable tobacco chemical composition analyzer is used to scan tobacco leaf samples and collect whole leaf spectral data information;

[0077] 2) Use a portable tobacco chemical composition analyzer to scan tobacco leaf samples and collect spectral data information of crushed leaves;

[0078] 3) The total sugar, reducing sugar, total alkaloids, potassium, chlorine and other chemical composition values ​​of the first-cured tobacco leaf samples were determined by continuous flow method. The statistical information of the chemical composition of the tobacco leaf samples is shown in Table 1;

[0079] Table 1 Statistical information of chemical composition of tobacco leaf samples

[0080] Table 1 Statistics information of chemical components in tobacco leaf samples

[0081]

[0082] 4) In order to avoid the high-frequency noise and baseline drift of near-infrared spectra caused by irrelevant factors, the spectral data were preprocessed using smoothing, derivative method, and standard normalization (SNV) processing methods to reduce the impact of noise and system errors on the spectrum;

[0083] 5) PLS modeling: Partial least squares (PLS) was selected for modeling. PLS combines three methods: multiple linear regression, canonical correlation analysis, and principal component analysis. The obtained model has a small prediction residual sum of squares and high prediction stability. PLS modeling was performed on the whole leaf spectral information of more than 300 samples. The results of internal cross-validation are shown in Table 2.

[0084] Table 2 Internal cross-validation results of near infrared model

[0085] Table 2 The cross validation results of NIR models

[0086]

[0087] 6) Transfer learning modeling: Based on the model obtained by PLS modeling, the association-ratio transfer learning method is used.

[0088] Assume there are models P and Q,

[0089] in , ,

[0090] Response variable , ,

[0091] Covariates , .

[0092] If models P and Q are both linear models, let:

[0093]

[0094] like ,and If exists, the model parameters have the following relationship:

[0095]

[0096] in .

[0097] In the present invention, it is assumed that the data obtained in the laboratory is model Q, and the spectral information obtained by scanning the whole leaf is , the spectral information obtained by scanning the broken leaves is The chemical composition of tobacco determined by continuous flow method is , according to PLS modeling, the model is obtained

[0098] , then according to the formula

[0099] Computational Model Parameters ,but , which is the prediction model we need. When predicting the chemical composition of tobacco leaves using near infrared spectroscopy, we only need to collect the spectral information of the whole tobacco leaf and bring it into the model P. The predicted values ​​of tobacco leaf chemical components can be obtained .

[0100] In the embodiment of the present invention, the accuracy evaluation criteria for the prediction value obtained in step S69 include the internal cross-prediction determination coefficient R 2The whole leaf spectra and broken leaf spectra of more than 300 samples were modeled, and the NIRS prediction models of total sugar, reducing sugar, total alkaloids, potassium, and chlorine after migration were established according to the formula. After internal cross-validation, the comparison with the results in Table 2 showed that the predicted values ​​had a good correlation with the corresponding chemical values. The prediction determination coefficient R2 and internal cross-validation root mean square (RMSECV) of all models were better than those before migration. The results are shown in Table 3.

[0101] Table 3 Internal cross-validation results of near infrared model

[0102] Table 3 The cross validation results of NIR models

[0103]

[0104] 7) Model testing: The internal cross-check of the model only shows the degree of consistency between the internal prediction value of the model and the chemical value. In order to check the adaptability of the model to unknown samples, an external cross-check must be performed. 87 tobacco leaf samples that did not participate in the modeling were randomly selected, and the post-migration mathematical model was used to predict the whole leaf. The results are shown in Table 4.

[0105] Table 4. Results of external verification of near infrared model

[0106] Table 4 The test validation results of NIR models

[0107]

[0108] As shown in Table 4, the external test results of the prediction models of each component show that the average values ​​of total sugar, reducing sugar, total alkaloids, potassium, and chlorine are 21.48%, 19.56%, 2.50%, 2.22%, and 0.33%, respectively, all of which belong to samples with normal contents, and the average relative errors are 2.68%, 2.77%, 4.39%, 8.95%, and 14.42%, and the prediction standard deviations are 3.11, 2.36, 0.62, 0.41, and 0.15, respectively. This shows that the prediction model has good prediction capabilities for the above five chemical components.

[0109] The experimental results show that the mathematical model established through transfer learning has a high degree of consistency in the predicted results and the results of continuous flow method determination. It is more stable than the previous models and can be directly transferred without resampling. It can be used for the analysis of the chemical components of tobacco leaves during the tobacco production process and meets the requirements for rapid detection of conventional chemical components of tobacco leaves.

[0110] Practice has proved that this method is an effective method for predicting tobacco chemical components. By using the transfer learning method to transfer the effective information in the broken leaf spectrum, the stability of the whole leaf spectrum measurement model is improved, and a new prediction model for total sugar, reducing sugar, total alkaloids, potassium, and chlorine content is established. The model has a high accuracy when predicting new samples, and the data obtained can meet the needs of tobacco production enterprises and meet the requirements for rapid detection of conventional chemical components of tobacco leaves.

[0111] The method provided by the present invention is different from the traditional model transfer method. Its purpose is to give full play to the operability of the whole leaf collection method compared with the broken leaf collection method. The model after migrating the effective information of the broken leaf spectrum improves the accuracy of the whole leaf spectrum measurement model, reduces the workload of re-collecting new samples and establishing models, and has good application effect.

[0112] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for optimizing the chemical composition model of tobacco leaf near-infrared spectra based on transfer learning, characterized in that: The optimization method comprises the following steps: Step S1, obtaining a near infrared spectrum of a whole tobacco leaf; Step S2, obtaining near infrared spectrum of broken tobacco leaves; Step S3, obtaining reference chemical composition values ​​by drying and grinding the tobacco leaves and testing them with a flow analyzer; Step S4, preprocessing the near infrared spectrum data obtained in step S1 and step S2; Step S5, according to the near-infrared spectral data of step S1 and step S2, based on the near-infrared spectral data of whole tobacco leaves and broken leaves, respectively establish chemical component analysis models for whole tobacco leaves and broken leaves; transfer the effective information of the broken tobacco leaf spectrum to the whole leaf spectrum model, establish a tobacco leaf near-infrared spectral chemical component optimization model based on transfer learning, and then use the above optimization model to perform component detection on the tobacco leaves to be tested; Step S6, transferring the broken leaf spectral information to the whole leaf spectral model through the association-ratio transfer learning method, so as to obtain an optimized model of the chemical components of the tobacco leaf near-infrared spectrum; detecting the chemical components of the tobacco leaf according to the optimized model, and obtaining the predicted values ​​of the chemical components of the tobacco leaf; In step S6, the optimization model of the chemical components of tobacco leaf near-infrared spectrum is implemented by the following steps: Step S61: firstly, a portable tobacco chemical component analyzer is used to scan a tobacco leaf sample to collect spectrum data information of the whole leaf; Step S62, using a portable tobacco chemical component analyzer to scan the crushed tobacco leaf samples and collect spectral data information of the crushed leaves; Step S63, using a continuous flow method to determine the chemical component value information of the total sugar, reducing sugar, total alkaloids, potassium, and chlorine content of the first-cured tobacco leaf sample; Step S64, pre-processing the spectral data using smoothing, derivative method, and standard normalized SNV processing methods in sequence to reduce the influence of noise and system errors on the spectrum; Step S65, select partial least squares PLS for modeling, the PLS prediction residual sum of squares is small and has high prediction stability; Step S66: According to the model obtained by PLS modeling in step S65, the association-ratio transfer learning method is used to calculate the parameter β of the optimization model according to the corresponding formula, so as to obtain the required optimization prediction model; The method for obtaining the predicted value of the tobacco leaf chemical composition is achieved by the following steps: Step S67, collecting spectrum information of the whole tobacco leaf; Step S68, preprocessing the sample spectrum data obtained in step S67; Step S69, substituting the data obtained in step S68 into the optimization model of the chemical components of tobacco leaves in near-infrared spectrum to obtain the predicted values ​​of the chemical components of tobacco leaves; In step S66, the transfer learning modeling method is: According to the model obtained by PLS modeling, the association-ratio transfer learning method is used. If there are models P and Q, where , , Response variable , , Covariates , ; If models P and Q are both linear models, let: like ,and exist, The model parameters have the following relationship: in , where Q is the laboratory data model, The spectral information obtained by scanning the whole leaf. The spectral information obtained from scanning the broken leaves, The chemical composition value of tobacco determined by continuous flow method; The model is obtained according to PLS modeling , Then according to the formula: , Calculate the parameters , Computational model This is the prediction model we need. When predicting the chemical composition of tobacco leaves using near infrared spectroscopy, we only need to collect the spectral information of the whole tobacco leaf and bring it into the model P. The predicted values ​​of tobacco leaf chemical components can be obtained ; The method for obtaining the predicted value of the chemical components of the tobacco leaves further includes converting the predicted value of the chemical components of the tobacco leaves obtained in step S69 into an expression understandable to the user according to a conversion rule and outputting the expression; The accuracy evaluation criteria for the prediction value obtained in step S69 include the internal cross-prediction determination coefficient R 2 Test and internal cross-root mean square RMSECV validation.

2. The method for optimizing the tobacco leaf near-infrared spectroscopy chemical composition model based on transfer learning according to claim 1, characterized in that: The reference chemical component values ​​include total sugar, reducing sugar, total alkaloids, total potassium and total chlorine values.

3. The method for optimizing the tobacco leaf near-infrared spectroscopy chemical composition model based on transfer learning according to claim 1, characterized in that: The near infrared spectrometer used to obtain the infrared spectrum in step S1 and step S2 is a YKS-H1 portable tobacco chemical composition analyzer.

Images