Non-contact prediction and inversion method for moisture content of tobacco sheet based on hyperspectral imaging technology under different coating rates

Abstract

The application discloses a tobacco sheet moisture content non-contact prediction and inversion method based on hyperspectral imaging technology under different coating rates, and belongs to the tobacco index detection field. In view of the problem that the spectral interference caused by the surface structure of the tobacco sheet is serious, and the process characteristics that the coating rate influences the detection precision, the application provides the following: collecting a near-infrared hyperspectral image of the tobacco sheet and extracting a spectrum of a region of interest; according to the coating rate, the samples are divided into a high-coating-rate group, a low-coating-rate group and a total sample set; the spectrum is preprocessed, such as first-order derivation, to enhance the moisture absorption characteristics; for each group of samples, a moisture prediction model is constructed based on a random forest regression algorithm; the performance of each model is evaluated through a comprehensive evaluation index, and the result is visualized. Through the combination of grouping modeling and targeted preprocessing, the application significantly improves the precision and robustness of moisture prediction under different coating rates, realizes rapid, nondestructive and intelligent detection of the moisture content of the tobacco sheet, and is suitable for online quality inspection and process optimization.

Description

Technical Field

[0001] This invention relates to the field of optical response detection applied to tobacco index detection. Specifically, it relates to a non-contact inversion method for the moisture content of tobacco sheets that combines hyperspectral imaging technology with machine learning algorithms. This method is particularly suitable for rapid, non-destructive, and accurate detection of moisture content in tobacco sheets under different coating rates. Background Technology

[0002] Tobacco sheets, also known as reconstituted tobacco, are one of the core raw materials in the tobacco industry. Their moisture content is a key indicator affecting storage and transportation safety, processing characteristics, sensory quality, and the final product grade. Excessive moisture content easily leads to mold growth, while insufficient moisture increases breakage, affecting combustion characteristics and taste. Therefore, accurate and rapid detection of the moisture content of tobacco sheets is of great significance for achieving automated raw material control and full-process quality traceability.

[0003] Currently, the conventional method for detecting moisture in tobacco sheets is oven drying. Although this method has high accuracy, it has inherent drawbacks such as being cumbersome to operate, time-consuming, damaging to the sample, and unable to achieve online detection. Furthermore, while online detection methods such as infrared moisture analyzers achieve non-contact measurement, they can only obtain point or small area information. For tobacco sheets with rough surfaces, complex wrinkles, and uneven thickness, they are difficult to represent the overall moisture distribution and are easily affected by material morphology and ambient light.

[0004] In recent years, hyperspectral imaging technology has been applied to quality inspection in the tobacco industry due to its advantage of combining images and spectra. This involves collecting hyperspectral images of fresh flue-cured tobacco leaves in specific wavelengths, extracting characteristic wavelengths using the CARS algorithm, and constructing a moisture prediction model using support vector regression. However, these industry solutions mainly focus on research objects, specific wavelengths, and data processing strategies, which differ from the tobacco sheets discussed in this invention. As a reconstituted industrial raw material, tobacco sheets have a much more complex physical structure than fresh tobacco leaves, including surface roughness, wrinkles, and thickness. This makes their spectral signals prone to severe baseline drift and scattering noise, making it difficult to achieve ideal prediction accuracy by directly applying methods used for fresh leaves.

[0005] Furthermore, existing technologies in the industry generally overlook the impact of coating rate, a key process parameter in tobacco sheet production, on moisture detection. Coating rate directly affects the density, porosity, and surface smoothness of the sheet, thereby altering the propagation path and absorption characteristics of light within it. Moreover, tobacco sheets with different coating rates exhibit significantly different spectral response characteristics; using existing technologies for prediction inevitably introduces systematic errors.

[0006] Therefore, the industry urgently needs to develop a hyperspectral moisture detection solution specifically designed for the complex structure of tobacco sheets and taking into full account the influence of coating rate, in order to solve specific problems such as poor adaptability and low accuracy. Summary of the Invention

[0007] In view of the above, the present invention aims to provide a non-contact prediction and inversion method for the moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates. The method effectively suppresses spectral interference caused by the complex structure of tobacco sheets through targeted preprocessing strategies, and innovatively introduces a coating rate grouping modeling strategy, which significantly improves the accuracy and robustness of moisture prediction under different process conditions.

[0008] The technical solution adopted in this invention is as follows: A non-contact prediction and inversion method for the moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates is provided, including:

[0009] Select tobacco sheet samples and, after removing defective products, determine the true moisture content of each sample.

[0010] The spectral image data of tobacco sheet samples in the near-infrared band was acquired using a hyperspectral imaging system. After black and white correction, the spectral mean of all pixels in the region of interest was extracted as the original spectral data of the sample, and the original spectral data was preprocessed.

[0011] Based on the coating rate of the tobacco sheet samples, the samples were divided into a high coating rate group, a low coating rate group, and an overall sample set;

[0012] For each group of samples, a moisture prediction model based on the random forest regression algorithm is constructed. The preprocessed spectral data is used as input and the true moisture content is used as output to train the tobacco sheet moisture content prediction model.

[0013] The coefficient of determination (R²), root mean square error (RMSE), and relative prediction deviation (RPD) were used as evaluation indicators to assess the impact of different preprocessing methods and prediction model combinations on the accuracy of water prediction, and the prediction results were presented in a visual manner.

[0014] In at least one of the possible implementations, the preprocessing includes at least one of first-derivative preprocessing and autoscaling preprocessing, for enhancing spectral features and suppressing baseline drift and scattering noise interference.

[0015] In at least one possible implementation, the grouping threshold for the coating rate is 39%: samples with a coating rate greater than 39% are classified into the high coating rate group, and samples with a coating rate less than 39% are classified into the low coating rate group.

[0016] In at least one of the possible implementations, the construction of the random forest regression model includes at least: automatically extracting spectral features related to the moisture content of tobacco sheets by integrating multiple decision trees, and outputting continuous predicted values ​​of moisture content.

[0017] In at least one of the possible implementations, while constructing the random forest regression model, partial least squares regression, adaptive boosting algorithm and artificial neural network model are also constructed as comparison models to evaluate the performance differences of different modeling methods.

[0018] In at least one possible implementation, the method further includes identifying key bands that contribute significantly to water content prediction based on the random forest regression model, including water absorption characteristic peaks around 1450 nm and 1940 nm.

[0019] In at least one possible implementation, the selection of tobacco sheet samples includes: reconstituted tobacco leaf samples that are structurally intact and uniformly colored, after manual screening and removal of damaged, moldy, and severely wrinkled materials; and,

[0020] The measured true moisture content values ​​for each sample include those calculated after drying at a constant temperature of 105℃ to constant weight using an oven method.

[0021] In at least one possible implementation, the near-infrared band ranges from 1000 nm to 2000 nm; and,

[0022] The acquisition parameters of the hyperspectral imaging system are set as follows: the object distance between the lens and the sample is 300 mm, the light source intensity is 210 W, the spectral resolution is 6.7 nm, the scanning speed is 21 mm / s, and the image size is 320 × 256 pixels.

[0023] In at least one of the possible implementations, the visualization method includes: a pseudo-color image generated based on the prediction results, which visually displays the spatial distribution and gradient changes of moisture in the tobacco sheet sample.

[0024] In at least one of the possible implementations, the results of model evaluation and optimization are displayed as a scatter plot to visually compare the consistency between model predictions and measured values ​​under different coating rate groups.

[0025] Compared to existing technologies, this invention specifically targets tobacco sheets, whose rough surface, complex wrinkles, and uneven thickness are key characteristics distinguishing them from other agricultural products such as fresh tobacco leaves, grains, and fruits. Addressing this characteristic, this invention proposes a sample grouping modeling strategy based on coating rate in tobacco sheet moisture detection. Unlike the usual assumption of uniform sample texture, coating rate, as a key process parameter, has a decisive impact on spectral characteristics. By dividing samples into high and low coating rate groups and modeling them separately, intra-group spectral heterogeneity is eliminated, allowing the model to more accurately learn the mapping relationship between moisture and spectrum within a specific coating rate range. Furthermore, the modeling method emphasizes a combination of preprocessing and grouping strategies based on multiple algorithms such as random forest regression. Leveraging the algorithmic ensemble learning mechanism, it can naturally handle high-dimensional, highly collinear spectral data without complex feature selection steps, automatically identifying and assigning higher weights to key bands, simplifying the modeling process, and exhibiting good immunity to overfitting. This makes it particularly suitable for tobacco material product applications with limited sample size and complex noise.

[0026] Furthermore, this invention, taking into account the characteristics of thin-film structures, systematically compared various processing methods and found that first-derivative preprocessing can effectively eliminate baseline drift and multiplicative scattering caused by complex surfaces, significantly enhance the signal-to-noise ratio of the 1450nm and 1940nm moisture characteristic absorption peaks, and lay the foundation for subsequent high-precision modeling.

[0027] Furthermore, experiments have shown that the model prediction performance (RPD up to 4.7972) under the grouped modeling strategy of this invention is significantly better than the non-grouped scheme.

[0028] In summary, this invention effectively solves the long-standing problems of inaccurate measurement and indistinguishable results in the moisture detection of tobacco sheets by using targeted object preprocessing, process parameter grouping, and robust ensemble learning algorithms, thus providing a reliable detection solution for the tobacco industry. Attached Figure Description

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings, wherein:

[0030] Figure 1 This is a schematic diagram of the overall process of the non-contact prediction and inversion method for the moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates provided in an embodiment of the present invention.

[0031] Figure 2 This is a schematic diagram of a tobacco sheet provided in an embodiment of the present invention;

[0032] Figure 3 The original reflectance spectra of different tobacco sheet samples in the 1000-2000 nm range provided in the embodiments of the present invention;

[0033] Figure 4 This is a spectrum of tobacco sheet after first-order derivative preprocessing provided in an embodiment of the present invention;

[0034] Figure 5 This is a pre-processed spectrum of tobacco sheet provided in an embodiment of the present invention after automatic scaling.

[0035] Figure 6 This is a scatter plot comparing the predicted and measured values ​​of the optimal random forest regression model for the high coating rate group provided in this embodiment of the invention.

[0036] Figure 7 This is a scatter plot comparing the predicted and measured values ​​of the optimal partial least squares regression model for the low coating rate group provided in this embodiment of the invention.

[0037] Figure 8 A scatter plot comparing the predicted and measured values ​​of the optimal random forest regression model for the overall sample set provided in this embodiment of the invention.

[0038] Figure 9 Spatial distribution diagram of moisture content of representative tobacco sheet samples with high coating rate provided in an embodiment of the present invention;

[0039] Figure 10 This is a spatial distribution diagram of moisture content in a representative tobacco sheet sample with low coating rate in an embodiment of the present invention. Detailed Implementation

[0040] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0041] This invention proposes an embodiment of a non-contact prediction and inversion method for the moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates. Specifically, as follows: Figure 1 As shown, it includes:

[0042] Step S1: Determine the true moisture content of the reconstituted tobacco sample and obtain the original near-infrared hyperspectral data of the sample;

[0043] This embodiment selects tobacco sheet samples provided by a supplier specializing in tobacco sheet production. To ensure data quality and sample consistency, the original samples were first manually screened, such as... Figure 2 As shown, after removing sheets with obvious appearance defects such as damage, mold, severe wrinkles or incomplete edges, a total of 282 representative tobacco sheet samples with complete structure, uniform color and good texture were retained.

[0044] Subsequently, the moisture content of all samples was determined. The oven drying method was used as the standard method, in which the samples were dried to constant weight at a constant temperature of 105℃, and the moisture percentage was calculated as a supervision label for subsequent model training.

[0045] Hyperspectral image acquisition was performed in a darkroom environment to eliminate ambient light interference. After the system warmed up for 30 minutes, the acquisition parameters were set as follows: lens-sample distance 300 mm, light source intensity 210 W, spectral resolution 6.7 nm, scanning speed 21 mm / s, and image size 320 × 256 pixels. The acquisition wavelength range was 1000-2000 nm. Before and after sample acquisition, a standard polytetrafluoroethylene white plate (reflectivity ~99.99%) was scanned, and a completely black image was obtained after the lens cap was closed for black and white correction.

[0046] In practice, the following correction formula can also be used:

[0047]

[0048] Where R is the corrected reflectivity, R0 is the original signal intensity, B is the full black calibration signal intensity, and W is the full white calibration signal intensity.

[0049] After acquisition, the region of interest for each tobacco sheet was extracted using an image segmentation algorithm, and the spectral data of all pixels within that region were averaged to obtain the original spectral feature curve of the sample, as shown below. Figure 3 As shown. From Figure 3 It can be seen that the original spectrum has obvious baseline drift and differences between samples, which will mask the characteristics of moisture absorption.

[0050] Step S2: Based on the moisture characteristics of tobacco sheets, preprocess the raw spectral data;

[0051] To reduce interference from illumination fluctuations and baseline drift, the raw spectral data is preprocessed. This invention focuses on the first derivative (1... st Two preprocessing methods are D) and Autoscale. First-order derivative processing can effectively eliminate baseline shift and linear tilt, highlighting spectral slope variations and absorption peak characteristics. For example... Figure 4 As shown in the comparison, it can be found that the first derivative processing significantly enhances the variation trends of several key bands, especially at the two typical moisture absorption peaks of 1450 nm and 1940 nm, where obvious valleys and abrupt slope changes occur, demonstrating stronger response sensitivity. These two bands correspond to the first and second vibrational-stretching combination frequencies of moisture (O–H bond absorption characteristics), respectively, and are important evidence for hyperspectral inversion of moisture.

[0052] Among them, the area around 1450 nm is a composite absorption band of the first-order water O–H bending vibration and stretching, which is more sensitive to samples with low water content; the area around 1940 nm is a strong absorption band jointly absorbed by free water and bound water, which responds more significantly to changes in samples with medium to high water content; the 1150–1250 nm region also has slight fluctuations, which may be related to structural water or intermolecular interactions. Although the absorption intensity is weak, it is still identifiable as a small peak in the derivative curve, which helps improve the accuracy of modeling.

[0053] In addition, refer to Figure 5 As illustrated, automatic scaling effectively compresses the spectral amplitude differences between samples, resulting in higher data consistency and facilitating the extraction of global features by machine learning models. In contrast, the first derivative is more advantageous in enhancing the gradient and slope changes of absorption features and is more suitable for characterizing the dynamic response of moisture in different bands. Therefore, it was selected as the dominant preprocessing method in this invention. Comprehensive analysis shows that, while maintaining the main absorption feature structure unchanged, first derivative processing significantly improves the signal-to-noise ratio of moisture-sensitive bands, effectively highlighting the learnable region of the model and providing a key foundation for high-precision modeling of moisture in reconstituted tobacco leaves using subsequent RF models.

[0054] Step S3: Divide the tobacco sheet samples into several groups based on the coating rate;

[0055] Based on the coating rate parameters from the tobacco sheet production records, all samples are classified. In some preferred embodiments of the invention, a coating rate of 39% can be set as the grouping threshold. Samples with a coating rate greater than 39% are classified into a high coating rate group, and samples with a coating rate less than 39% are classified into a low coating rate group. All samples are then merged into a total sample set. This grouping operation is one of the key concepts of this embodiment, aiming to eliminate spectral heterogeneity caused by differences in process parameters and make model learning more focused.

[0056] Step S4: Construct multiple models using different groupings and conduct comparative evaluations;

[0057] For the high coating rate group, low coating rate group, and overall sample set, four algorithms—Random Forest (RF), Partial Least Squares Regression (PLS), Adaptive Boosting Algorithm (AdaBoost), and Artificial Neural Network (ANN)—were used to construct moisture prediction models. Specifically, the preprocessed data for each group were randomly divided into training and test sets at a ratio of approximately 4:1. Model performance was comprehensively evaluated using the coefficient of determination (R²), root mean square error (RMSE), and relative prediction deviation (RPD). The RPD value is the ratio of the standard deviation of the prediction performance to the RMSE. Generally, an RPD > 3.0 indicates that the model has excellent predictive ability.

[0058] (1) Analysis of results for the high coating rate group:

[0059] As shown in Table 1, in the high coating rate group, the best performing model combination is the first derivative + random forest (RF). Its test set determination coefficient R² reaches 0.9547, root mean square error RMSEP is 0.1510, and relative prediction deviation RPD is as high as 4.7972. This indicates that after first derivative preprocessing, the RF model can fully extract information from the moisture-sensitive bands, resulting in extremely high prediction accuracy. The corresponding scatter plot is shown below. Figure 6 As shown, the data points are closely distributed around the y=x line. In contrast, although the ANN model fits better on the calibration set, its performance fluctuates greatly on the test set, posing a risk of overfitting. Table 1 below is a comparison of the prediction performance indicators of different models and preprocessing combinations for the high coating rate group in the embodiments of this invention.

[0060] Table 1

[0061]

[0062] (2) Analysis of results for the low coating rate group:

[0063] As shown in Table 2, the best model for the low coating rate group was first derivative + partial least squares regression (PLS), with an R² of 0.9200 and an RPD of 3.3843. This indicates that in the low coating rate samples, the linear relationship between spectral information and moisture content is more prominent, and the linear model PLS can also achieve excellent results. Nevertheless, the performance of the RF model (RPD 2.8745) remains robust. Figure 7 The fitting effect of the optimal model group is shown. Table 2 below is a comparison table of the prediction effect indicators of different models and preprocessing combinations for the low coating rate group in the embodiments of the present invention.

[0064] Table 2

[0065]

[0066] (3) Analysis of overall sample set results:

[0067] As shown in Table 3 below, in the ungrouped overall sample, the best model is autoscaling + Random Forest (RF), with an RPD of 3.8924. This indicates that as sample complexity increases, the nonlinear fitting ability and robustness of the RF model allow it to maintain high performance. Meanwhile, autoscaling preprocessing is the most effective in this group, suggesting that data standardization is a more critical preprocessing step for samples containing all process parameters. Figure 8 The optimal model set is shown in Table 3 below, which compares the prediction performance of different models and preprocessing combinations for the overall sample set in this embodiment of the invention.

[0068] Table 3

[0069]

[0070] Step S5: Visualize and analyze the moisture content of different groups predicted by the model.

[0071] To visually demonstrate the model's spatial prediction capabilities, Figure 9 and Figure 10 Three image representations of six groups of tobacco sheet samples with different coating rates are presented: (1) pseudo-color image based on prediction results; (2) corresponding hyperspectral image grayscale distribution; and (3) actual visible light image. From the pseudo-color image, it can be observed that the model successfully distinguishes tobacco sheets with different moisture ranges, and their color gradients show a clear increasing trend. The predicted value range from 6.3% to 10.3% is well covered, indicating that the present invention has high discrimination and prediction consistency at the image level. Furthermore, comparison with the visible light image also shows a high degree of consistency between the model's prediction results and the actual color depth trend, further verifying the reliability of its visualization inversion capability. This indicates that the method of the present invention not only predicts accurately in numerical terms but also achieves spatial visualization inversion of moisture at the image level, providing richer information for process optimization.

[0072] In summary, this invention, through targeted first-derivative preprocessing, coating rate grouping strategy, and robust diversified modeling, successfully achieves high-precision, non-contact prediction of the moisture content of tobacco sheets under different coating rates. For example, in the high coating rate group, it achieves an excellent performance with an RPD value exceeding 4.7, fully demonstrating that the technical solution of this invention provides a reliable technical means for online quality inspection and index control in the tobacco industry, and has extremely high application value and promotion prospects.

[0073] In this invention, when directional terms are mentioned, they are relative concepts based on the embodiments. Furthermore, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0074] The above description of the structure, features, and effects of the present invention is based on the embodiments shown in the figures. However, the above are only preferred embodiments of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred methods can be reasonably combined and matched by those skilled in the art to form a variety of equivalent solutions without departing from or changing the design concept and technical effects of the present invention. Therefore, the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A non-contact prediction and inversion method for the moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates, characterized in that, include: Select tobacco sheet samples and, after removing defective products, determine the true moisture content of each sample. The spectral image data of tobacco sheet samples in the near-infrared band was acquired using a hyperspectral imaging system. After black and white correction, the spectral mean of all pixels in the region of interest was extracted as the original spectral data of the sample, and the original spectral data was preprocessed. Based on the coating rate of the tobacco sheet samples, the samples were divided into a high coating rate group, a low coating rate group, and an overall sample set; For each group of samples, several moisture prediction models, including at least the random forest regression algorithm, are constructed. The preprocessed spectral data is used as input and the true moisture content is used as output to train the tobacco sheet moisture content prediction model. The coefficient of determination (R²), root mean square error (RMSE), and relative prediction deviation (RPD) were used as evaluation indicators to assess the impact of different preprocessing methods and prediction model combinations on the accuracy of water prediction, and the prediction results were presented in a visual manner.

2. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates as described in claim 1, characterized in that, The preprocessing includes at least one of first derivative preprocessing and autoscaling preprocessing, used to enhance spectral features and suppress baseline drift and scattering noise interference.

3. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates as described in claim 1, characterized in that, The grouping threshold for the coating rate is 39%: samples with a coating rate greater than 39% are classified into the high coating rate group, and samples with a coating rate less than 39% are classified into the low coating rate group.

4. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates as described in claim 1, characterized in that, The construction of a random forest regression model includes at least the following steps: by integrating multiple decision trees, automatically extracting spectral features related to the moisture content of tobacco sheets, and outputting continuous predicted values ​​of moisture content.

5. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates as described in claim 4, characterized in that, While constructing the random forest regression model, we also constructed partial least squares regression, adaptive boosting algorithm and artificial neural network model as comparison models to evaluate the performance differences of different modeling methods.

6. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates as described in claim 4, characterized in that, The method also includes identifying key bands that contribute significantly to water content prediction based on the random forest regression model, including water absorption characteristic peaks around 1450 nm and 1940 nm.

7. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates as described in claim 1, characterized in that, The selected tobacco sheet samples include: reconstituted tobacco leaf samples that have been manually screened, and after removing damaged, moldy, and severely wrinkled materials, retaining those with intact structure and uniform color; and, The measured true moisture content values ​​for each sample include those calculated after drying at a constant temperature of 105℃ to constant weight using an oven method.

8. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates as described in claim 1, characterized in that, The near-infrared band ranges from 1000 nm to 2000 nm; and, The acquisition parameters of the hyperspectral imaging system are set as follows: the object distance between the lens and the sample is 300 mm, the light source intensity is 210 W, the spectral resolution is 6.7 nm, the scanning speed is 21 mm / s, and the image size is 320 × 256 pixels.

9. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates according to any one of claims 1 to 8, characterized in that, The visualization method includes: a pseudo-color image generated based on the prediction results, which intuitively displays the spatial distribution and gradient changes of moisture in tobacco sheet samples.

10. The non-contact prediction and inversion method for moisture content of tobacco sheets based on hyperspectral imaging technology under different coating rates according to any one of claims 1 to 8, characterized in that, The results of model evaluation and optimization are presented as scatter plots to visually compare the consistency between model predictions and measured values ​​under different coating rate groups.

Images