Method for monitoring tobacco mellowing speed based on near infrared spectrum and application thereof

Abstract

The invention provides a method for monitoring the alcoholization speed of tobacco leaves based on a near infrared spectrum and application thereof, and the method comprises the following steps: carrying out near infrared spectrum detection on the tobacco leaves to be detected, and substituting the detection result into a regression model to obtain the alcoholization speed of the tobacco leaves to be detected. The method provided by the invention overcomes the defect that only the tobacco mellowing degree can be statically predicted in the prior art, the time change rate of chemical and sensory indexes is innovatively defined as the mellowing speed label, and the regression model is constructed in combination with the dynamic characteristics of the near infrared spectrum, so that the rapid and accurate monitoring of the tobacco mellowing dynamic speed is realized for the first time; and the blank in dynamic analysis of the alcoholization process in the prior art is filled.

Description

Technical Field

[0001] This invention belongs to the field of tobacco aging monitoring technology, specifically relating to a method for monitoring the aging rate of tobacco leaves based on near-infrared spectroscopy and its application. Background Technology

[0002] Tobacco aging is a crucial step in tobacco processing that determines its quality. Through natural or artificial control, it induces a series of complex physiological and biochemical reactions within the tobacco leaves, gradually reducing irritation and off-flavors while improving aroma quality and harmony. Quality control during aging relies on precise monitoring of its progress and rate; therefore, establishing efficient and real-time aging status monitoring technology has always been a key research focus in the tobacco industry.

[0003] Traditional methods for monitoring tobacco aging primarily rely on offline chemical analysis and sensory evaluation. Chemical analysis uses instruments such as high-performance liquid chromatography (HPLC) and gas chromatography (GC) to detect the content of key chemical indicators in tobacco leaves, such as glucose, fructose, nicotine, and total nitrogen, to determine the degree of aging. Sensory evaluation involves professionals subjectively assessing the quality of tobacco leaves based on multiple dimensions, including aroma, off-odors, and irritation. However, these methods have significant limitations: firstly, chemical analysis is cumbersome and time-consuming, and the sample processing may damage the original state of the tobacco leaves, making continuous monitoring impossible; secondly, sensory evaluation results are heavily influenced by human experience, are highly subjective, and struggle to quantify the dynamic changes during the aging process.

[0004] With the development of spectral analysis technology, near-infrared spectroscopy (NIRS) has been increasingly applied to the monitoring of tobacco aging due to its advantages of speed, non-destructiveness, and simultaneous detection of multiple components. Existing NIRS-based methods primarily focus on predicting the static degree of aging, that is, assessing the current aging state of tobacco leaves by establishing a correlation model between spectral data and the content of chemical indicators or sensory scores at a specific time point. Although these static prediction methods have improved monitoring efficiency to some extent, they have not yet overcome the limitations of "static state description" and cannot capture the dynamic evolution of the aging process. In fact, the core of tobacco aging lies in "change"—the rate of increase or decrease of chemical indicator content over time and the speed of improvement in sensory quality directly determine the direction of aging process optimization.

[0005] Therefore, in view of the shortcomings of existing technologies that can only statically assess the degree of aging and cannot capture dynamic speed information, developing a technical method that can accurately quantify the aging speed of tobacco leaves and realize dynamic monitoring throughout the process is of great significance for improving the controllability of tobacco leaf aging quality and optimizing the production process. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a method for monitoring the aging rate of tobacco leaves based on near-infrared spectroscopy and its application. The method provided by this invention overcomes the limitation of existing technologies that can only statically predict the degree of tobacco aging. It innovatively defines the "time-varying rate of change of chemical and sensory indicators" as the aging rate label, and constructs a regression model based on the dynamic characteristics of near-infrared spectroscopy. This achieves, for the first time, rapid and accurate monitoring of the dynamic aging rate of tobacco leaves, filling the gap in the dynamic analysis of the aging process in existing technologies.

[0007] To achieve this objective, the present invention adopts the following technical solution:

[0008] On the one hand, the present invention provides a method for monitoring the aging rate of tobacco leaves based on near-infrared spectroscopy, the method comprising the following steps:

[0009] The tobacco leaves to be tested were subjected to near-infrared spectroscopy, and the detection results were substituted into a regression model to obtain the aging rate of the tobacco leaves to be tested.

[0010] The regression model is obtained by a method including the following steps:

[0011] (1) Select tobacco leaves at different aging stages, collect their monitoring indicators, establish a fitting curve of aging stage-monitoring indicator data and calculate the first derivative to obtain the rate of change of monitoring indicators over time, which is recorded as aging speed (i.e. the change range of monitoring indicators per unit time, such as the rate of change of sensory evaluation indicators in minutes / time).

[0012] (2) Near-infrared spectroscopy detection was performed on tobacco leaves at different aging stages and data was collected. After data processing, the obtained near-infrared spectral data was used as input feature X and aging rate was used as output target Y to construct a near-infrared spectroscopy-dynamic rate association dataset. The dataset was divided into a training set (for model building) and a test set (real samples for model testing) for model training, and a regression model that can dynamically output the predicted values ​​of each aging rate was obtained.

[0013] The monitoring indicators include sensory evaluation indicators.

[0014] The above method innovatively defines the "time change rate of chemical and sensory indicators" as the aging rate label, realizing a technological leap from the "static state description" of conventional methods to the "dynamic rate monitoring" of the present invention. It overcomes the shortcomings of existing technologies that can only statically predict the degree of tobacco aging. By combining the dynamic characteristics of near-infrared spectroscopy to construct a regression model, it achieves rapid and accurate monitoring of the dynamic rate of tobacco aging for the first time, filling the gap in the dynamic analysis of the aging process in existing technologies.

[0015] Preferably, the different alcoholization stages in step (1) include five different alcoholization stages.

[0016] Preferably, the monitoring indicators also include any one or a combination of at least two of the following: fructose content, glucose content, chlorogenic acid content, or rutin content.

[0017] Preferably, the fitting curve in step (1) is obtained by fitting a fourth-order polynomial.

[0018] Preferably, the tobacco leaves are pretreated before collection in step (2), and the pretreatment includes baking the tobacco leaves and then crushing them.

[0019] Preferably, the baking temperature is 45-55℃ and the time is 1.5-2.5 h. The temperature can be 45℃, 46℃, 47℃, 48℃, 49℃, 50℃, 51℃, 52℃, 53℃, 54℃ or 55℃, etc., and the time can be 1.5 h, 1.6 h, 1.7 h, 1.8 h, 1.9 h, 2 h, 2.1 h, 2.2 h, 2.3 h, 2.4 h or 2.5 h, etc., but is not limited to the values ​​listed above. Other unlisted values ​​within the above range are also applicable.

[0020] Preferably, the scanning band for the near-infrared spectroscopy detection is 700-2500 nm.

[0021] Preferably, the data processing method in step (2) includes Savitzky-Golay smoothing and standard normal variable transformation.

[0022] Preferably, the model training in step (2) uses the default network structure of the TabPFN algorithm.

[0023] On the other hand, the present invention also provides the application of the method described above in the aging process of tobacco leaves.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] This invention provides a method for monitoring the aging rate of tobacco leaves based on near-infrared spectroscopy. By innovatively defining the "time change rate of chemical and sensory indicators" as the aging rate label, it achieves a technological leap from the "static state description" of conventional methods to the "dynamic rate monitoring" of this invention. It overcomes the shortcomings of existing technologies that can only statically predict the degree of tobacco aging. By combining the dynamic characteristics of near-infrared spectroscopy to construct a regression model, it achieves rapid and accurate monitoring of the dynamic rate of tobacco aging for the first time, filling the gap in the dynamic analysis of the aging process in existing technologies. Attached Figure Description

[0026] Figure 1 This is a comparison of the spectra before and after pretreatment in Example 1;

[0027] Figure 2 These are the correlation curves of the sensory evaluation index for the Wenshan C2FA sample, where the left side is the fitted curve and the right side is the first derivative curve of the fitted curve;

[0028] Figure 3 This is a graph showing the prediction results of the TabPFN regression model for the "near-infrared spectroscopy-glucose" C2FA sample in Guiyang, Chenzhou, Hunan.

[0029] Figure 4 This is a graph showing the prediction results of the "near-infrared spectroscopy-fructose" TabPFN regression model for the C2FA sample in Guiyang, Chenzhou, Hunan.

[0030] Figure 5 This is a graph showing the prediction results of the TabPFN regression model for the "near-infrared spectroscopy-sensory assessment" of the C2FA sample in Guiyang, Chenzhou, Hunan.

[0031] Figure 6 This is a graph showing the prediction effect of the TabPFN regression model for the C2FA sample in Guiyang, Chenzhou, Hunan Province, based on "near-infrared spectroscopy-chlorogenic acid".

[0032] Figure 7 This is a graph showing the prediction effect of the TabPFN regression model for the "near-infrared spectroscopy-rutin" of the C2FA sample in Guiyang, Chenzhou, Hunan. Detailed Implementation

[0033] The technical solution of the present invention will be further illustrated below through specific embodiments. Those skilled in the art should understand that the embodiments described are merely illustrative of the present invention and should not be construed as limiting the invention in any way.

[0034] Example 1

[0035] This embodiment provides a method for monitoring the aging rate of tobacco leaves based on near-infrared spectroscopy. The specific steps are as follows:

[0036] S01: Select 12 representative tobacco leaf samples, covering 5 consecutive aging stages; the sample collection for each aging stage follows the principles of strict randomness and representativeness. Two boxes of aging tobacco leaves are randomly selected from the aging warehouse, and 2 kg of samples are taken from each box of tobacco leaves using the five-point sampling method to avoid data distortion due to sampling bias.

[0037] S02: Glucose and fructose were detected in tobacco leaf samples using high-performance liquid chromatography (HPLC) for quantitative analysis. A Yuexu Ultimate XB-NH2 column (4.6 mm × 150 mm, 3 μm particle size) was used. The mobile phase was acetonitrile-water (78:22 v / v), the flow rate was controlled at 1 mL / min, the column temperature was maintained at 40℃, and the injection volume was 10 μL. An evaporative light detector was used during the detection process, with the drift tube temperature set at 85℃ and the nitrogen flow rate controlled at 2 L / min to ensure the accuracy and repeatability of the detection results. Sensory evaluation of tobacco leaf samples was conducted by a panel of five professional evaluators certified by the State Tobacco Monopoly Administration. Following the "Sensory Evaluation Methods for Tobacco and Tobacco Products" (GB / T 16447-2004), the panel independently scored 29 sub-indicators across four categories: aroma characteristics, quality characteristics, comfort characteristics, and style characteristics. For example, aroma characteristics included 11 sub-indicators such as caramel sweetness, light sweetness, and woody sweetness; quality characteristics included 7 sub-indicators such as aroma quantity, aroma quality, and grassy / unpleasant aromas. The average values ​​were summed to obtain the total sensory evaluation score for each sample, ensuring the objectivity and reliability of the results.

[0038] Chlorogenic acid and rutin were detected in tobacco leaf samples using high-performance liquid chromatography-ultraviolet (HPLC-UV) for quantitative analysis. The chromatographic conditions were as follows: a SHIMADZUShil-pack GIS column (4.6 mm × 250 mm, 5 μm) was used; mobile phase A was water-methanol-glacial acetic acid (89:10:1, V / V / V); mobile phase B was water-methanol-glacial acetic acid (10:89:1, V / V / V); the gradient elution program is shown in Table 1; the flow rate was 1 mL / min; the column temperature was 30℃; the injection volume was 10 μL; and the UV detector wavelength was 340 nm. After grinding and mixing the tobacco leaf samples, 0.5 g was accurately weighed and placed in a 50 mL centrifuge tube. 20 mL of 50% methanol aqueous solution was added, and the mixture was shaken well and ultrasonically extracted for 20 min. The extract was filtered through a 0.22 μm aqueous filter membrane, and the filtrate was used as the analyte. The analytes were then analyzed under the chromatographic conditions described above.

[0039] Table 1 Elution procedure for chlorogenic acid and rutin by high performance liquid chromatography-ultraviolet detection.

[0040]

[0041] The results are shown in Table 2:

[0042] Table 2 Sample Information and Test Data

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] S03: 2 kg of tobacco leaves from each batch were dried in a 50℃ constant temperature oven for 2 hours. After drying, they were immediately placed in sealed bags to cool. The cooled tobacco leaves were then removed, pulverized using a traditional Chinese medicine pulverizer, and passed through a 0.25 mm sieve. Ten parallel small samples were prepared for each aging stage of each tobacco leaf sample for subsequent spectral acquisition repeatability verification. A PerkinElmer Lambda 1050+ near-infrared spectrometer was used, with the following spectral acquisition parameters: scanning band of 700-2500 nm, data interval of 4 nm, and integration time of 0.08 s for both the photomultiplier tube and the photoelectric sensor. Near-infrared spectra were acquired from three different locations for each of the ten parallel small samples at each aging stage to eliminate the influence of sample inhomogeneity and instrument fluctuations. The average of the three spectra for each small sample was then taken as the representative spectral data for that small sample. The spectral data were organized according to the aging stage sequence to form a dynamic spectral dataset that continuously changes with the aging process.

[0050] S04: Near-infrared spectral data were preprocessed using Savitzky-Golay (SG) smoothing (smoothing window = 15) and Standard Normal Variable Transform (SNV). A comparison of the spectra before and after preprocessing is shown below. Figure 1 , Figure 1 A represents the unprocessed raw near-infrared spectrum, with a slight drift in the visible spectral baseline and significant fluctuations in local bands due to noise interference. Figure 1 B is the near-infrared spectrum after SG smoothing, which significantly improves baseline flatness, greatly reduces noise interference, and significantly optimizes the smoothness of the spectral curve and the identification of characteristic peaks. Figure 1 C represents the near-infrared spectrum after combined preprocessing with SG smoothing and SNV. This not only maintains the low noise advantage of smoothing but also further unifies the spectral baseline levels of different samples, making the differences in spectral characteristics at different alcoholization stages more prominent.

[0051] The glucose content, fructose content, sensory evaluation data, chlorogenic acid content, and rutin content of different tobacco leaves at various aging stages were processed separately. First, a dynamic dataset of "aging stage - index content" was constructed using the polyfit function of the NumPy library. The fitting parameters were set to polynomial degree = 4 and numerical stability parameter (rcond) = None (the function automatically adapts to the data range to ensure numerical stability). Then, the first derivative of the fitted curve was calculated, and the derivative result was defined as the "aging speed" label of the corresponding stage, thus completing the transformation from static index data to dynamic speed quantification. Figure 2 The figures show the correlation curves of the sensory evaluation index for the Wenshan C2FA sample. Subplot A is the fitted curve of the index, and subplot B is the first derivative curve of the fitted curve.

[0052] Next, model building and monitoring were carried out. First, near-infrared spectral data preprocessed with SG smoothing and SNV were used as input features (X), and the "maturation rate" label obtained by fitting the first derivative of the curve corresponding to each aging stage was used as the output target (Y), thus constructing a "near-infrared spectrum-dynamic rate" association dataset. To ensure good generalization ability of the model, a "hierarchical random partitioning" strategy was adopted, dividing the dataset into a training set (for model building) and a test set (real samples) in a 7:3 ratio. This partitioning method ensures that data from different aging stages of each tobacco leaf sample are covered in both the training and test sets, avoiding model bias due to uneven data distribution. Then, the TabPFN algorithm was used to train the model on the training set. Leveraging its efficient few-shot learning capability, a regression model that can dynamically output the predicted rates for each aging stage was obtained.

[0053] Further analysis of the predictive performance of each indicator model reveals: Figure 3 This image shows the prediction performance of the TabPFN regression model for the "near-infrared spectroscopy-glucose" C2FA sample in Guiyang, Chenzhou, Hunan. The goodness of fit of the training set is R0. 2 =1.000, Mean Absolute Error (MAE) = 0.024, Test Set R 2 = 0.996, MAE = 0.457, the scatter points closely surround the fitted line, demonstrating the model's excellent predictive ability for glucose alcoholization rate; Figure 4 This is a graph showing the prediction performance of the "near-infrared spectroscopy-fructose" TabPFN regression model for the C2FA sample in Guiyang, Chenzhou, Hunan. The training set is R. 2 = 1.000, MAE = 0.050, test set R 2 = 0.986, MAE = 0.963, also showing a high degree of fit, indicating that the model can effectively capture the correlation between near-infrared spectroscopy and fructose alcoholization rate; Figure 5This image shows the prediction performance of the TabPFN regression model for the "near-infrared spectroscopy-sensory absorption" of the C2FA sample in Guiyang, Chenzhou, Hunan. The training set is R. 2 = 1.000, MAE = 0.118, test set R 2 = 0.999, MAE = 0.546; Figure 6 This image shows the prediction performance of the TabPFN regression model for the "near-infrared spectroscopy-chlorogenic acid" method for the C2FA sample in Guiyang, Chenzhou, Hunan. The training set is R. 2 = 1.000, MAE = 0.008, test set R 2 = 0.999, MAE = 0.047; Figure 7 This image shows the prediction performance of the TabPFN regression model for the "near-infrared spectroscopy-rutin" C2FA sample in Guiyang, Chenzhou, Hunan. The training set is R. 2 = 1.000, MAE = 0.030, test set R 2 = 1.000, MAE = 0.038; further verifying the accuracy of the model in predicting the aging rate through sensory evaluation. By preprocessing tobacco leaf samples at unknown aging stages according to the workflow, collecting spectra, and inputting them into the corresponding index model, the predicted aging rate value of the index can be obtained quickly. By continuously collecting spectral data at different time points and inputting them into the model, the rate change trend can also be tracked in real time. Finally, by combining the models of various indices, the dynamic monitoring and tracking of the aging rate of tobacco leaves throughout the entire process can be achieved.

[0054] The glucose, fructose, and sensory evaluation of the remaining samples R 2 The MAE results are shown in Tables 3-7. Tables 3, 4, 5, 6, and 7 present the performance of the TabPFN model for different tobacco leaf samples under five categories of indicators: glucose, fructose, sensory evaluation, chlorogenic acid, and rutin. From the training set indicators, the training set R... 2 The MAE values ​​are mostly close to or reach 1.000, and the training set MAE is generally small, indicating that the model fits the training data extremely well and can accurately learn the data patterns; on the test set, the R-value of most samples is close to or reaches 1.000. 2 The model's performance is at a relatively high level, and the MAE on the test set is relatively controllable, demonstrating good generalization ability and the ability to effectively predict the aging rate of different tobacco leaf samples. However, the R-value on some test sets is low. 2 The fluctuations in MAE reflect differences in the intrinsic relationship between the spectral characteristics and aging rate of tobacco leaves from different origins and varieties, resulting in varying model adaptability. Overall, the TabPFN model demonstrates good reliability and accuracy in predicting tobacco aging rate.

[0055] Table 3. Glucose TabPFN Model Indicators

[0056]

[0057] Table 4. Fructose TabPFN Model Indicators

[0058]

[0059] Table 5 Sensory Evaluation TabPFN Model Indicators

[0060]

[0061] Table 6. Chlorogenic acid TabPFN model indices

[0062]

[0063] Table 7. Model Indicators of Rutin TabPFN

[0064]

[0065] As can be seen from the above, the method provided by the present invention can quickly and accurately predict the aging rate of tobacco leaves, filling the gap in the dynamic analysis of the aging process in the existing technology.

[0066] The applicant declares that this invention illustrates the method for monitoring the aging rate of tobacco leaves based on near-infrared spectroscopy and its application through the above embodiments. However, this invention is not limited to the above embodiments, meaning that this invention does not necessarily rely on the above embodiments for implementation. Those skilled in the art should understand that any improvements to this invention, equivalent substitutions of raw materials for the product of this invention, addition of auxiliary components, and selection of specific methods, etc., all fall within the protection and disclosure scope of this invention.

[0067] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0068] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

Claims

1. A method for monitoring the aging rate of tobacco leaves based on near-infrared spectroscopy, characterized in that, The method includes the following steps: The tobacco leaves to be tested were subjected to near-infrared spectroscopy, and the detection results were substituted into a regression model to obtain the aging rate of the tobacco leaves to be tested. The regression model is obtained by a method including the following steps: (1) Select tobacco leaves at different aging stages, collect their monitoring indicators, establish a fitting curve of aging stage-monitoring indicator data and calculate the first derivative to obtain the rate of change of monitoring indicators over time, which is denoted as aging speed. (2) Near-infrared spectroscopy detection was performed on tobacco leaves at different aging stages and data was collected. After data processing, the obtained near-infrared spectral data was used as input feature X and aging rate was used as output target Y to construct a near-infrared spectroscopy-dynamic rate association dataset. The dataset was divided into training set and test set for model training to obtain a regression model that can dynamically output the predicted values ​​of each aging rate. The monitoring indicators include sensory evaluation indicators.

2. The method according to claim 1, characterized in that, Step (1) describes five different alcoholization stages.

3. The method according to claim 1 or 2, characterized in that, The monitoring indicators also include any one or a combination of at least two of the following: fructose content, glucose content, chlorogenic acid content, or rutin content.

4. The method according to any one of claims 1-3, characterized in that, The fitting curve in step (1) is obtained by fitting a fourth-order polynomial.

5. The method according to any one of claims 1-4, characterized in that, Step (2) involves pre-treating the tobacco leaves before collection. The pre-treatment includes heating the tobacco leaves and then crushing them.

6. The method according to claim 5, characterized in that, The baking temperature is 45-55℃, and the time is 1.5-2.5 h.

7. The method according to any one of claims 1-6, characterized in that, The scanning band for the near-infrared spectroscopy detection is 700-2500 nm.

8. The method according to any one of claims 1-7, characterized in that, The data processing methods described in step (2) include Savitzky-Golay smoothing and standard normal variable transformation.

9. The method according to any one of claims 1-8, characterized in that, The model training in step (2) uses the default network structure of the TabPFN algorithm.

10. The application of the method according to any one of claims 1-9 in the aging process of tobacco leaves.

Images