Tobacco leaf slitting method based on multi-view weight learning and storage medium

Abstract

The application discloses a tobacco slitting method based on multi-view weight learning and a storage medium, obtains a hyperspectral image of target batch tobacco and performs division; an effective area of the tobacco is obtained by using a threshold segmentation method, the tobacco is divided into a set number of subareas with the same longitudinal length, characteristic spectra of the subareas corresponding to the tobacco are calculated, and a tobacco spectrum database is constructed; spectrum data in the tobacco spectrum database is subjected to band division, different band combinations are obtained, and a multi-view tobacco spectrum database is constructed; a semi-supervised clustering algorithm model is constructed based on multi-view weight and similarity learning, the semi-supervised clustering algorithm model is trained through the multi-view tobacco spectrum database, and a semi-supervised clustering algorithm model with an optimized target function is obtained; and a tobacco segmentation result is obtained through the semi-supervised clustering algorithm model. Through construction of the multi-view of the tobacco, weights are allocated to each view, differences between different tobaccos are accurately quantified, and the tobacco slitting effect is ensured.

Description

Technical Field

[0001] This invention belongs to the field of tobacco leaf cutting technology, specifically relating to a tobacco leaf cutting method and storage medium based on multi-view weight learning. Background Technology

[0002] In the cigarette production system, precise control of tobacco leaf quality is the core foundation for ensuring the consistency of the sensory quality of the final product. Even within the same origin, grade, and even the same leaf, leaves from different locations (such as the tip, middle, base, and sides of the midrib) exhibit significant differences in key chemical components (such as nicotine, sugars, and total nitrogen), physical structure (thickness, toughness), and even combustion characteristics due to variations in maturity, light exposure, and nutrient accumulation. These naturally occurring locational differences pose a serious challenge to cigarette formulation: failure to achieve precise locational slicing will disrupt the homogeneity of the formulation, leading to batch-to-batch fluctuations in smoke flavor and affecting the stability of the cigarette's smoking experience and the precision of quality control. Therefore, developing an intelligent tobacco leaf locational slicing technology to accurately classify and process leaves from different characteristic locations has become a key technological requirement for improving the homogeneity of cigarette production and quality assurance capabilities.

[0003] Traditional sorting methods based on manual experience or ordinary imaging methods struggle to meet the aforementioned requirements in terms of accuracy and stability. In recent years, hyperspectral imaging technology has provided a powerful solution. Compared to conventional RGB images or single-point spectral detection, the greatest advantage of hyperspectral imaging lies in its ability to simultaneously acquire continuous, narrow-band (typically hundreds of channels) spectral information for each spatial pixel. Due to differences in their intrinsic chemical composition and physical structure, tobacco leaves from different locations exhibit unique and sensitive absorption and reflection characteristics in specific spectral bands (such as the visible and near-infrared regions). For example, thicker cell structures at the leaf base may exhibit strong absorption in specific short-wave infrared bands, while areas rich in sugar accumulation may show characteristic absorption peaks in specific near-infrared bands. This "image-spectrum integration" characteristic of hyperspectral data allows it to capture deeper, more subtle spectral differences in leaves from different locations that are directly related to their intrinsic quality, thus providing a refined spatial distribution characterization of tobacco leaf quality that surpasses human visual recognition and ordinary image analysis. The introduction of hyperspectral technology has significantly enhanced the potential for identifying the locational quality of tobacco leaves. The formation of different quality zones in tobacco leaves is the result of the combined effects of multiple biochemical processes and physical structures, and modeling from a single perspective is difficult to cover all its attribute information.

[0004] Currently available tobacco leaf slicing technologies mainly fall into two categories: one is slicing methods based on the distribution patterns of chemical components, such as the Chinese invention patent with authorization announcement number CN113080506 B. This method divides green tobacco leaves into multiple segments along the main vein, using the weight ratio of reducing sugar to total sugar as an indicator, and achieves the substitution of slicing segments with graded tobacco leaves through cluster analysis. While this method can effectively improve the utilization rate of green tobacco leaves, it relies on chemical detection methods, resulting in complex sample pretreatment, long detection cycles, and significant tobacco leaf loss. Furthermore, it only focuses on sugar indicators and fails to comprehensively consider multi-dimensional attributes such as nicotine and physical thickness, limiting its ability to comprehensively characterize differences in tobacco leaf quality.

[0005] Another type of slitting method is based on physical thickness distribution, such as the Chinese invention patent application CN114376260 A. This method uses a thickness gauge to detect the thickness changes of different sections of the tobacco leaf and combines it with Fisher's optimal segmentation method to determine the optimal slitting scheme. Although this method has the advantages of simple operation, high detection efficiency, and no loss of tobacco leaves, it only relies on a single physical index (thickness) for slitting and ignores the differences in important quality attributes such as chemical composition and combustion characteristics. This may result in significant fluctuations in the sensory quality and processing behavior of the slit tobacco leaves, making it difficult to meet the stringent requirements of high-end cigarettes for raw material homogenization.

[0006] A new multi-view weighted learning method for tobacco leaf segmentation is needed to address the technical problem that existing segmentation methods, which only process data from a single viewpoint, are unable to decouple the differences in various physicochemical factors, thus failing to accurately quantify the differences among multiple factors affecting the quality of tobacco leaves in different locations. Summary of the Invention

[0007] The purpose of this invention is to provide a tobacco leaf slicing method based on multi-view weight learning, which solves the technical problems of single slicing view and poor slicing effect.

[0008] Another objective of this invention is to provide a computer-readable storage medium.

[0009] The technical solution of this invention to solve its technical problem is as follows:

[0010] A tobacco leaf segmentation method based on multi-view weight learning includes the following steps:

[0011] S1: Obtain hyperspectral images of the target batch of tobacco leaves, and classify the hyperspectral images into first-grade tobacco leaf images, second-grade tobacco leaf images, and third-grade tobacco leaf images according to the quality classification standards of tobacco leaves;

[0012] S2: Use threshold segmentation to remove the background, veins, and shadows of tobacco leaves from the first-level, second-level, and third-level tobacco leaf images to obtain the effective area of ​​the tobacco leaf. Based on the size of the effective area, divide the tobacco leaf into a set number of sub-regions with the same longitudinal length along the direction from the leaf tip to the leaf base. Calculate the average spectrum of all pixels in the middle area of ​​each sub-region and use the average spectrum as the feature spectrum of the corresponding sub-region of the tobacco leaf to construct a tobacco leaf spectral database.

[0013] S3: By dividing the spectral data in the tobacco leaf spectral database into bands, different band combinations are obtained, and different band combinations are used to construct a multi-view tobacco leaf spectral database.

[0014] S4: Construct a semi-supervised clustering algorithm model based on multi-view weight and similarity learning. The objective function of the semi-supervised clustering algorithm model includes a similarity matrix, a weight regularization term, a similarity matrix regularization term, and a spectral clustering embedding term. Train the semi-supervised clustering algorithm model using a multi-view tobacco leaf spectral database to obtain a semi-supervised clustering algorithm model with an optimized objective function.

[0015] S5: Input the hyperspectral image of the tobacco leaf to be segmented into a semi-supervised clustering algorithm model. The semi-supervised clustering algorithm model divides the hyperspectral image of the tobacco leaf to be segmented into a set number of sub-regions with the same longitudinal length along the direction from the leaf tip to the leaf base based on the optimized objective function. This serves as the initial result for tobacco leaf quality segmentation. The initial result is then iteratively optimized, and the optimized tobacco leaf segmentation result is output.

[0016] Preferably, the objective function in step S4 is:

[0017] ;

[0018] The constraints are as follows: , , ; This represents a similarity matrix between spectral images of tobacco leaves. Indicates the first Weight of each perspective Represents the embedding term matrix, Represents the distance between samples. Indicates the first The first tobacco leaf Spectral images from various perspectives Indicates the first The first tobacco leaf Spectral images from various perspectives Describes the first similar matrix. The first tobacco leaf spectral image and the first Similarity between individual tobacco leaf spectral images This represents the weight regularization term. This represents the weight regularization parameter. Represents the weight vector. , This represents the regularization term for similarity matrices. This represents the regularization parameter for similarity matrices. This represents the spectral clustering embedding term. Represents the trace of a matrix. express The transpose of the matrix; Represents the Laplace matrix, , Let the diagonal matrix be represented by its first... The diagonal elements are represented as n represents the number of elements. express The transpose of the matrix, , Labels indicating known categories of tobacco leaves; This indicates an unknown category label for tobacco leaves. This indicates the dimension of each spectral segment. , Indicates the number of spectral bands in tobacco leaves. This indicates the perspective of the tobacco leaves.

[0019] Preferably, the training of the semi-supervised clustering algorithm model in step S4 specifically includes the following steps:

[0020] S4.1: For similarity matrices Weight vector Embedding Item Matrix Perform initialization.

[0021] S4.2: Update the embedding matrix The known category label of tobacco leaves is passed to the unknown category label of tobacco leaves; the objective function is... ,Will Divide into blocks to obtain , ;

[0022] S4.3: Update the similarity matrix between tobacco leaf spectral images Adaptive learning and optimization of the similarity relationship between tobacco leaf samples, with the objective function being: ,

[0023] ,

[0024] ,

[0025] ,

[0026] ,

[0027] in, Represents the first Lagrange multiplier. This represents the number of nearest neighbors in the tobacco leaf spectrum; Indicates the quantity of tobacco leaves; This represents the weighted multi-view feature distance. Indicates the spatial distance of the embedding;

[0028] S4.4: Update multi-view weights It automatically evaluates and assigns weights to different perspectives, with the objective function being... ,

[0029] ,

[0030] , , Indicates the second Lagrange multiplier. Indicates the first Total distance cost from each perspective;

[0031] S4.5: Repeat steps S4.2, S4.3, and S4.4 until the objective converges, resulting in a new objective function.

[0032] Preferably, step S5 specifically includes the following steps:

[0033] S5.1: Input the hyperspectral image of the tobacco leaf to be segmented into a semi-supervised clustering algorithm model. The semi-supervised clustering algorithm model divides the hyperspectral image of the tobacco leaf to be segmented into a set number of sub-regions with the same longitudinal length along the direction from leaf tip to leaf base based on the optimized objective function, as the initial result of tobacco leaf quality segmentation.

[0034] S5.2: Calculate the center point of each sub-region, search for similar points in the region where the center point is located and its adjacent regions, and assign the similar pixels to the region where the center point has the smallest comprehensive distance based on the combined distance between the region center point and all similar pixels within its search range. The specific calculation formula is as follows:

[0035] ,

[0036] ,

[0037] ,

[0038] in, Indicates the center point of the region and the first point within its search range. Multi-view weighted spectral distance of each pixel Represents spatial coordinate distance. To determine the overall distance of pixel allocation, A parameter representing the relative importance of spectral distance and spatial distance. ;

[0039] S5.3: Based on the current allocation results of similar pixels, recalculate the center point of each sub-region;

[0040] S5.4: Repeat steps S5.2 and S5.3 to update the allocation results of similar pixels until the loop stopping condition is met, and output the optimized tobacco leaf segmentation result.

[0041] Preferably, the loop stopping condition is: the number of loops reaches 50.

[0042] Preferably, the number set in steps S2 and S5 is the same as the number of viewpoints.

[0043] Preferably, the set quantity and number of viewpoints are 5.

[0044] Preferably, the quality classification standard for the tobacco leaves is an expert experience standard.

[0045] Preferably, the first-level tobacco leaf image, the second-level tobacco leaf image, and the third-level tobacco leaf image correspond to the three categories of good, medium, and poor tobacco leaves, respectively.

[0046] A computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it causes the device on which the computer-readable storage medium resides to execute a tobacco leaf cutting method based on multi-view weight learning.

[0047] The beneficial effects of this invention are as follows: By acquiring hyperspectral images of the target batch of tobacco leaves, the hyperspectral images are divided into first-grade, second-grade, and third-grade tobacco leaf images according to the quality classification standards of the tobacco leaves; the effective region of the tobacco leaves is obtained using a threshold segmentation method, and the tobacco leaves are divided into a set number of sub-regions with the same vertical length. The characteristic spectra of the corresponding sub-regions of the tobacco leaves are calculated to construct a tobacco leaf spectral database; then, by dividing the spectral data in the tobacco leaf spectral database into bands, different band combinations are obtained, and different band combinations are used to construct a multi-view tobacco leaf spectral database; a semi-supervised clustering algorithm model is constructed based on multi-view weights and similarity learning. The objective function of the supervised clustering algorithm model includes a similarity matrix, a weight regularization term, a similarity matrix regularization term, and a spectral clustering embedding term. A semi-supervised clustering algorithm model is trained using a multi-view tobacco leaf spectral database to obtain a semi-supervised clustering algorithm model with an optimized objective function. The hyperspectral image of the tobacco leaf to be segmented is input into the semi-supervised clustering algorithm model. Based on the optimized objective function, the semi-supervised clustering algorithm model divides the hyperspectral image of the tobacco leaf to be segmented into a predetermined number of sub-regions with the same vertical length along the leaf tip to leaf base direction. This serves as the initial result for tobacco leaf quality segmentation. The initial result is iteratively optimized, and the optimized tobacco leaf segmentation result is output. By constructing multiple views of the tobacco leaf and assigning weights to each view, the differences between tobacco leaves in different locations are accurately quantified, ensuring the effectiveness of tobacco leaf segmentation. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating the tobacco leaf cutting method of the present invention;

[0049] Figure 2 This is an image showing the tobacco leaf being divided into five regions using the tobacco leaf cutting method of this invention.

[0050] Figure 3 This is a multi-view weighted map of B2F, C2F, and C3F tobacco in the Kunming area in this embodiment of the invention;

[0051] Figure 4 This is a multi-view weighted map of cloud smoke B2F, C2F, and C3F in Chuxiong region in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0053] like Figure 1 As shown, this invention discloses a tobacco leaf segmentation method based on multi-view weight learning, comprising the following steps:

[0054] S1: Obtain hyperspectral images of the target batch of tobacco leaves. Classify these images into Grade 1, Grade 2, and Grade 3 images according to the tobacco leaf quality classification standards. The quality classification standards are based on expert experience, with Grade 1, Grade 2, and Grade 3 images corresponding to good, medium, and poor categories, respectively. The target batch can be a specific variety and grade of tobacco leaves from a specific region, such as Kunming Yunyan C2F.

[0055] S2: Using threshold segmentation, the background, veins, and shadows of tobacco leaves are removed from the primary, secondary, and tertiary tobacco leaf images to obtain the effective area of ​​the tobacco leaf. Based on the size of the effective area, the tobacco leaf is divided into a predetermined number of sub-regions with the same vertical length along the direction from the leaf tip to the leaf base. The average spectrum of all pixels in the middle region of each sub-region is calculated, and this average spectrum is used as the feature spectrum of the corresponding sub-region of the tobacco leaf to construct a tobacco leaf spectral database; the specific database is as follows: ,Right now They represent the 1st to the 2nd. A dataset from multiple perspectives. The size is set to 5, such as... Figure 2 As shown, there are 5 areas marked with different colors.

[0056] S3: By dividing the spectral data in the tobacco leaf spectral database into bands, different band combinations are obtained, and different band combinations are used to construct a multi-view tobacco leaf spectral database.

[0057] S4: Construct a semi-supervised clustering algorithm model based on multi-view weights and similarity learning. The objective function of the semi-supervised clustering algorithm model includes a similarity matrix, a weight regularization term, a similarity matrix regularization term, and a spectral clustering embedding term. Train the semi-supervised clustering algorithm model using a multi-view tobacco leaf spectral database to obtain a semi-supervised clustering algorithm model with an optimized objective function.

[0058] The objective function is: ;

[0059] The constraints are as follows: , , ; This represents a similarity matrix between spectral images of tobacco leaves. Indicates the first Weight of each perspective Represents the embedding term matrix, Represents the distance between samples. Indicates the first The first tobacco leaf Spectral images from various perspectives Indicates the first The first tobacco leaf Spectral images from various perspectives Describes the first similar matrix. The first tobacco leaf spectral image and the first Similarity between individual tobacco leaf spectral images This represents the weight regularization term. This represents the weight regularization parameter. Represents the weight vector. , This represents the regularization term for similarity matrices. This represents the regularization parameter for similarity matrices. This represents the spectral clustering embedding term. Represents the trace of a matrix. express The transpose of the matrix; Represents the Laplace matrix, , Let the diagonal matrix be represented by its first... The diagonal elements are represented as n represents the number of elements. express The transpose of the matrix, , Labels indicating known categories of tobacco leaves; This indicates an unknown category label for tobacco leaves. This indicates the dimension of each spectral segment. , Indicates the number of spectral bands in tobacco leaves. The number of angles represents the tobacco leaf, and the number of angles is the same as the number of angles, that is, the number of angles is 5.

[0060] Training a semi-supervised clustering algorithm model includes the following steps:

[0061] S4.1: For similarity matrices Weight vector Embedding Item Matrix Perform initialization.

[0062] S4.2: Update the embedding matrix The known category label of tobacco leaves is passed to the unknown category label of tobacco leaves; the objective function is... ,Will Divide into blocks to obtain , ;

[0063] S4.3: Update the similarity matrix between tobacco leaf spectral images Adaptive learning and optimization of the similarity relationship between tobacco leaf samples, with the objective function being: ,

[0064] ,

[0065] ,

[0066] ,

[0067] ,

[0068] in, Represents the first Lagrange multiplier. This represents the number of nearest neighbors in the tobacco leaf spectrum; Indicates the quantity of tobacco leaves; This represents the weighted multi-view feature distance. Indicates the spatial distance of the embedding;

[0069] S4.4: Update multi-view weights It automatically evaluates and assigns weights to different perspectives, with the objective function being... ,

[0070] ,

[0071] , , Indicates the second Lagrange multiplier. Indicates the first Total distance cost from each perspective;

[0072] S4.5: Repeat steps S4.2, S4.3, and S4.4 until the objective converges, resulting in a new objective function.

[0073] S5: Input the hyperspectral image of the tobacco leaf to be segmented into a semi-supervised clustering algorithm model. The semi-supervised clustering algorithm model divides the hyperspectral image of the tobacco leaf to be segmented into a set number of sub-regions with the same longitudinal length along the direction from the leaf tip to the leaf base based on the optimized objective function. This serves as the initial result for tobacco leaf quality segmentation. The initial result is then iteratively optimized, and the optimized tobacco leaf segmentation result is output.

[0074] Specifically, the following steps are included:

[0075] S5.1: Input the hyperspectral image of the tobacco leaf to be segmented into a semi-supervised clustering algorithm model. The semi-supervised clustering algorithm model divides the hyperspectral image of the tobacco leaf to be segmented into a set number of sub-regions with the same longitudinal length along the direction from leaf tip to leaf base based on the optimized objective function, as the initial result of tobacco leaf quality segmentation.

[0076] S5.2: Calculate the center point of each sub-region, search for similar points in the region where the center point is located and its adjacent regions, and assign the similar pixels to the region where the center point has the smallest comprehensive distance based on the combined distance between the region center point and all similar pixels within its search range. The specific calculation formula is as follows:

[0077] ,

[0078] ,

[0079] ,

[0080] in, Indicates the center point of the region and the first point within its search range. Multi-view weighted spectral distance of each pixel Represents spatial coordinate distance. To determine the overall distance of pixel allocation, A parameter representing the relative importance of spectral distance and spatial distance. ;

[0081] S5.3: Based on the current allocation results of similar pixels, recalculate the center point of each sub-region;

[0082] S5.4: Repeat steps S5.2 and S5.3, updating the allocation results of similar pixels until the loop stopping condition is met, and output the optimized tobacco leaf segmentation result. The loop stopping condition is: the number of loops reaches 50.

[0083] Example 1:

[0084] Experiments were conducted using Yunyan B2F, C2F, and C3F tobacco leaves from Kunming and Chuxiong regions. Each grade contained 60 tobacco leaves, with 20 leaves each of good, medium, and poor quality. Specific data are shown in Table 1.

[0085]

[0086] The hyperspectral images of tobacco leaves captured in the experiment covered a spectral range of 1000 nm to 2400 nm, comprising a total of 220 bands. These bands were divided into six groups, or six spectral perspectives. The first five perspectives each contained 36 bands, while the last perspective contained 40 bands. The multi-view weights for Kunming and Chuxiong obtained using the above method are as follows: Figure 3 , Figure 4As shown in the figure, the weights of the six perspectives for B2F, C2F, and C3F grade tobacco leaves in Kunming and Chuxiong regions exhibit the same trend. This indicates that the six perspectives contribute consistently to the quality quantification of different grades of tobacco leaves when differentiating their quality. The tobacco leaf slicing results obtained through this method ensure consistency, providing a more robust and interpretable slicing basis for tobacco leaf slicing.

[0087] A computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it causes the device on which the computer-readable storage medium resides to execute a tobacco leaf cutting method based on multi-view weight learning.

[0088] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

Claims

1. A tobacco leaf segmentation method based on multi-view weight learning, characterized in that: Includes the following steps: S1: Obtain hyperspectral images of the target batch of tobacco leaves, and classify the hyperspectral images into first-grade tobacco leaf images, second-grade tobacco leaf images, and third-grade tobacco leaf images according to the quality classification standards of tobacco leaves; S2: Use threshold segmentation to remove the background, veins, and shadows of tobacco leaves from the first-level, second-level, and third-level tobacco leaf images to obtain the effective area of ​​the tobacco leaf. Based on the size of the effective area, divide the tobacco leaf into a set number of sub-regions with the same longitudinal length along the direction from the leaf tip to the leaf base. Calculate the average spectrum of all pixels in the middle area of ​​each sub-region and use the average spectrum as the feature spectrum of the corresponding sub-region of the tobacco leaf to construct a tobacco leaf spectral database. S3: By dividing the spectral data in the tobacco leaf spectral database into bands, different band combinations are obtained, and different band combinations are used to construct a multi-view tobacco leaf spectral database. S4: Construct a semi-supervised clustering algorithm model based on multi-view weight and similarity learning. The objective function of the semi-supervised clustering algorithm model includes a similarity matrix, a weight regularization term, a similarity matrix regularization term, and a spectral clustering embedding term. Train the semi-supervised clustering algorithm model using a multi-view tobacco leaf spectral database to obtain a semi-supervised clustering algorithm model with an optimized objective function. S5: Input the hyperspectral image of the tobacco leaf to be segmented into a semi-supervised clustering algorithm model. The semi-supervised clustering algorithm model divides the hyperspectral image of the tobacco leaf to be segmented into a set number of sub-regions with the same longitudinal length along the direction from the leaf tip to the leaf base based on the optimized objective function. This serves as the initial result for tobacco leaf quality segmentation. The initial result is then iteratively optimized, and the optimized tobacco leaf segmentation result is output.

2. The tobacco leaf slicing method based on multi-view weight learning according to claim 1, characterized in that: The objective function in step S4 is: ； The constraints are as follows: , , ; This represents a similarity matrix between spectral images of tobacco leaves. Indicates the first Weight of each perspective Represents the embedding term matrix, Represents the distance between samples. Indicates the first The first tobacco leaf Spectral images from various perspectives Indicates the first The first tobacco leaf Spectral images from various perspectives Describes the first similar matrix. The first tobacco leaf spectral image and the first Similarity between individual tobacco leaf spectral images This represents the weight regularization term. This represents the weight regularization parameter. Represents the weight vector. , This represents the regularization term for similarity matrices. This represents the regularization parameter for similarity matrices. This represents the spectral clustering embedding term. Represents the trace of a matrix. express The transpose of the matrix; Represents the Laplace matrix, , Let the diagonal matrix be represented by its first... The diagonal elements are represented as n represents the number of elements. express The transpose of the matrix, , Labels indicating known categories of tobacco leaves; This indicates an unknown category label for tobacco leaves. This indicates the dimension of each spectral segment. , Indicates the number of spectral bands in tobacco leaves. This indicates the perspective of the tobacco leaves.

3. The tobacco leaf slicing method based on multi-view weight learning according to claim 2, characterized in that: Step S4, which involves training the semi-supervised clustering algorithm model, specifically includes the following steps: S4.1: For similarity matrices Weight vector Embedding Item Matrix Perform initialization. S4.2: Update the embedding matrix The known category label of tobacco leaves is passed to the unknown category label of tobacco leaves; the objective function is... ,Will Divide into blocks to obtain , ; S4.3: Update the similarity matrix between tobacco leaf spectral images Adaptive learning and optimization of the similarity relationship between tobacco leaf samples, with the objective function being: , , , ， , in, Represents the first Lagrange multiplier. This represents the number of nearest neighbors in the tobacco leaf spectrum; Indicates the quantity of tobacco leaves; This represents the weighted multi-view feature distance. Indicates the spatial distance of the embedding; S4.4: Update multi-view weights It automatically evaluates and assigns weights to different perspectives, with the objective function being... , , , , Indicates the second Lagrange multiplier. Indicates the first Total distance cost from each perspective; S4.5: Repeat steps S4.2, S4.3, and S4.4 until the objective converges, resulting in a new objective function.

4. The tobacco leaf slicing method based on multi-view weight learning according to claim 3, characterized in that: Step S5 specifically includes the following steps: S5.1: Input the hyperspectral image of the tobacco leaf to be segmented into a semi-supervised clustering algorithm model. The semi-supervised clustering algorithm model divides the hyperspectral image of the tobacco leaf to be segmented into a set number of sub-regions with the same longitudinal length along the direction from leaf tip to leaf base based on the optimized objective function, as the initial result of tobacco leaf quality segmentation. S5.2: Calculate the center point of each sub-region, search for similar points in the region where the center point is located and its adjacent regions, and assign the similar pixels to the region where the center point has the smallest comprehensive distance based on the combined distance between the region center point and all similar pixels within its search range. The specific calculation formula is as follows: , , , in, Indicates the center point of the region and the first point within its search range. Multi-view weighted spectral distance of each pixel Represents spatial coordinate distance. To determine the overall distance of pixel allocation, A parameter representing the relative importance of spectral distance and spatial distance. ; S5.3: Based on the current allocation results of similar pixels, recalculate the center point of each sub-region; S5.4: Repeat steps S5.2 and S5.3 to update the allocation results of similar pixels until the loop stopping condition is met, and output the optimized tobacco leaf segmentation result.

5. The tobacco leaf slicing method based on multi-view weight learning according to claim 4, characterized in that: The loop stops when the number of iterations reaches 50.

6. The tobacco leaf slicing method based on multi-view weight learning according to claim 1, characterized in that: The number set in steps S2 and S5 is the same as the number of viewpoints.

7. The tobacco leaf slicing method based on multi-view weight learning according to claim 6, characterized in that: The set quantity and number of viewpoints are 5.

8. The tobacco leaf slicing method based on multi-view weight learning according to claim 1, characterized in that: The quality classification standard for the tobacco leaves is based on expert experience.

9. The tobacco leaf slicing method based on multi-view weight learning according to claim 7, characterized in that: The first-level tobacco leaf images, second-level tobacco leaf images, and third-level tobacco leaf images correspond to the three categories of good, medium, and poor tobacco leaves, respectively.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the device containing the computer-readable storage medium to perform the method of claims 1-9.

Images