A Fresh Tobacco Leaf Position Recognition Method Based on Machine Vision

Abstract

The present invention discloses a method for identifying the positions of fresh tobacco leaves based on machine vision, which includes: establishing a library of harvested pictures of leaves at different positions of fresh tobacco leaves, and placing a standard scale beside the fresh tobacco leaves during harvesting and photographing; preprocessing the pictures, and using the Canny operator to extract the contour edge curve of the flue-cured tobacco in the preprocessed pictures; constructing a circumscribed rectangle of the leaf contour; extracting leaf feature parameters based on the circumscribed rectangle of the leaf contour, taking the number of pixel points of the line connecting the two points with the largest distance on the leaf contour as the maximum leaf length, taking the number of pixel points of the line connecting the two points that are perpendicular to the leaf length and have the farthest distance on the leaf contour as the maximum leaf width, and establishing a discrimination model for the positions of fresh tobacco leaves according to the maximum leaf length, the maximum leaf width and the aspect ratio. The present invention realizes the automatic machine identification of the positions of fresh tobacco leaves, solves the problem of low identification efficiency caused by manual identification of the positions of fresh tobacco leaves, and lays a foundation for automatically matching baking process parameters to achieve intelligent baking.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, in particular to the automatic recognition and classification of fresh tobacco leaves, and specifically to a method for recognizing the parts of fresh tobacco based on machine vision. Background Technique

[0002] The part is an important factor affecting the baking process parameters of tobacco leaves. After harvesting, the setting of the baking process parameters of fresh tobacco is judged by baking technicians based on characteristics such as parts and maturity, and the subjective differences are relatively large. The quality of the tobacco leaves after baking is uneven. With the development of the Internet of Things intelligent baking technology, the automatic baking of tobacco leaves based on image recognition will gradually replace manual tobacco baking.

[0003] In recent years, there have been great progress in the research on plant classification based on image analysis. Filipa et al. used a method of setting thresholds with a hybrid model to segment the image background, and achieved precise segmentation of plant images based on the feature ranking method; Dong Benzhi et al. detected corner points in leaf images based on the Freeman chain code method, and the calculation accuracy of leaf area and leaf perimeter was significantly improved; Dong Hongxia et al. based on 7 relative geometric features and texture features such as narrowness and rectangularity, and used a BP neural network to classify leaves, with an identification accuracy rate of 98.4%; Wei Lei et al. selected multiple feature parameters and used an SVM classifier to identify 4 types of leaves, with an accuracy rate of 95.8%; Lukic et al. used Hu invariant moments and the LBP algorithm to extract features, and SVM was used as a classifier for plant classification and recognition; Li Yang et al. proposed a plant leaf recognition algorithm based on morphological features and used KNN-SVM to classify and recognize leaves; Qi Zhang et al. evaluated 10 types of classifiers and found that in the recognition of leaves with different features, random forests and logistic regression have higher accuracy and stability. These research results can verify that different machine vision methods have the characteristics of simplicity, high efficiency, and non-destructiveness compared with methods such as manual detection, gas phase spectroscopy, and liquid phase spectroscopy.

[0004] At present, great progress has also been made in the research on the method of machine vision in the field of tobacco recognition, but it mainly focuses on the grading of tobacco leaves after baking and the recognition of the maturity of tobacco leaves. For example, a method for intelligent recognition and grade determination of flue-cured tobacco RGB images in an open environment disclosed in the Chinese patent with the publication number CN110415181A on November 5, 2019. And most of the existing technologies use image features such as color and texture, and there are few reports on the recognition of the parts of fresh tobacco leaves. In the flue-cured tobacco harvesting link, there are certain differences in the appearance morphological characteristics and dry matter accumulation of flue-cured tobacco in different parts, resulting in great differences in the baking characteristics of tobacco leaves. Correctly recognizing the parts of tobacco leaves and matching reasonable process parameters is an important part of improving the baking quality of tobacco leaves. Summary of the Invention

[0005] To overcome the deficiencies of the above-mentioned existing technologies, the present invention provides a method for identifying the positions of fresh tobacco leaves based on machine vision. This method utilizes machine vision technology to extract the morphological characteristic parameters of tobacco leaves and adopts machine learning methods to establish a classification and recognition model to achieve automatic and accurate identification of the positions of fresh tobacco leaves.

[0006] The present invention is realized through the following technical solutions:

[0007] A method for identifying the positions of fresh tobacco leaves based on machine vision, the method comprising:

[0008] Establish a leaf collection picture library for different positions of fresh tobacco leaves, and place a standard scale beside the fresh tobacco leaves when taking pictures for collection;

[0009] Preprocess the pictures in the leaf collection picture library, and use the canny operator to extract the contour edge curve of the flue-cured tobacco from the preprocessed pictures;

[0010] Construct a circumscribed rectangle of the leaf contour based on the extracted contour edge curve of the flue-cured tobacco;

[0011] Extract the leaf characteristic parameters based on the circumscribed rectangle of the leaf contour. Take the number of pixel points of the line connecting the two points with the largest distance on the leaf contour as the maximum leaf length, take the number of pixel points of the line connecting the two points that are perpendicular to the leaf length and have the farthest distance on the leaf contour as the maximum leaf width, and calculate the leaf width ratio according to the maximum leaf length and the maximum leaf width;

[0012] According to the maximum leaf length, the maximum leaf width and the length-width ratio, establish a discrimination model for the positions of fresh tobacco leaves as: position = a + b × maximum leaf length + c × maximum leaf width + d × length-width ratio.

[0013] The above technical solution extracts the leaf characteristic parameters through the contour edge curve of the flue-cured tobacco and establishes a discrimination model between the parameters of the fresh tobacco leaves and the positions according to the extracted leaf characteristic parameters, realizing the direct conversion from the parameters of the fresh tobacco leaves to the positions. This technical solution only needs to obtain the image of the fresh tobacco leaves, extract the leaf characteristic parameters and input them into the discrimination model for the positions of the fresh tobacco leaves, and then the discrimination result of the position can be directly obtained, solving the problem of low efficiency caused by manual discrimination before baking and then inputting the position information to match the process parameters, which is beneficial to the digital customization of the baking process.

[0014] The above technical solution is applied to the intelligent baking process. Since there is no longer manual intervention, but machine vision is used to accurately discriminate the positions of the flue-cured tobacco, and then match the best baking process parameters, the accuracy of using machine vision to discriminate the positions of the flue-cured tobacco can be improved to achieve the purpose of improving the baking quality.

[0015] Since the standard scale is used as a reference when taking pictures of the leaves, in the obtained images, the maximum leaf length, the maximum leaf width and the leaf width ratio can be calculated according to the pixel ratio of the standard scale.

[0016] As a further technical solution, the method further includes: inputting the maximum leaf length and leaf width ratio of the fresh tobacco leaves into the fresh tobacco leaf position discrimination model, outputting the position of the current leaf, and matching the corresponding baking process according to the current position.

[0017] The output end of the fresh tobacco leaf position discrimination model can be connected to the baking controller, and the baking controller automatically matches the corresponding baking process according to the position of the fresh tobacco leaves, realizing an intelligent operation process for determining the position of the fresh tobacco leaves and judging the baking process.

[0018] As a further technical solution, the position discrimination model of fresh tobacco leaves of different varieties is calculated according to the stepwise regression equation. By obtaining a large amount of data on the maximum leaf length, maximum leaf width and leaf width ratio of fresh tobacco leaves of the same variety, the position discrimination model of fresh tobacco leaves of this variety is calculated using the stepwise regression equation.

[0019] For different tobacco leaf varieties, different fresh tobacco leaf position discrimination models can be established to facilitate the rapid determination of the positions of fresh tobacco leaves of different varieties and the judgment of the baking process.

[0020] As a further technical solution, when calculating the stepwise regression equation, the variance inflation coefficient of each leaf characteristic parameter is statistically calculated, and the leaf characteristic parameters with a variance inflation coefficient greater than the preset value are excluded.

[0021] When calculating the stepwise regression equation using the maximum leaf length, maximum leaf width and leaf width ratio, there will be problems with the low significance of some characteristic parameters. Therefore, by statistically calculating the variance inflation coefficient of each characteristic parameter and excluding the non-significant characteristic parameters according to the variance inflation coefficient, the calculation amount of the discrimination process is reduced on the premise of ensuring the discrimination accuracy of the fresh tobacco leaves.

[0022] As a further technical solution, when photographing the fresh tobacco leaf image, the harvested fresh tobacco leaves are laid flat in the center of a black background cloth, a standard scale is placed beside, the camera is fixed at a preset distance from the ground, and the lens is adjusted to be perpendicular to the ground for image acquisition.

[0023] After setting up the camera, background cloth and standard scale, for newly harvested fresh tobacco leaves, only need to place the fresh tobacco leaves at the shooting position to automatically collect their images, and automatically output the position of the current fresh tobacco leaves based on machine vision recognition, and then match the corresponding baking process according to the position of the fresh tobacco leaves, realizing the rapid and accurate judgment of the fresh tobacco leaf baking process and improving work efficiency.

[0024] As a further technical solution, when establishing the leaf harvesting picture library for different parts of fresh tobacco leaves, the number of pictures of upper leaves (leaf positions 16 - 21), middle leaves (leaf positions 8 - 15), and lower leaves (leaf positions 1 - 7) is not less than 100 each. Such a setting can meet the training and testing requirements of the fresh tobacco leaf part discrimination model, enabling the constructed model to meet the accuracy requirements.

[0025] As a further technical solution, label the upper leaves, middle leaves, and lower leaves in the leaf harvesting picture library respectively, and randomly select the training set and test set in a preset proportion for training and testing the constructed fresh tobacco leaf part discrimination model. Before constructing the fresh tobacco leaf part discrimination model, first obtain the leaf samples for model construction, and then have professional personnel who have been engaged in flue-cured tobacco production work for a long time conduct part discrimination and labeling on the leaf samples to obtain the leaves with part labels for model training and testing.

[0026] As a further technical solution, the preprocessing of the collected images further includes: image grayscale, Gaussian filtering for smoothing, binaryzation of the grayscale image, and morphological processing to achieve the separation of the leaf from the background. This technical solution achieves the separation of the leaf from the background through a series of processes on the originally collected leaf images, facilitating the subsequent extraction of the leaf contour curve.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] (1) The present invention extracts the leaf feature parameters through the flue-cured tobacco contour edge curve, and establishes the fresh tobacco leaf parameter and part discrimination model based on the extracted leaf feature parameters, realizing the direct conversion from fresh tobacco leaf parameters to parts. This method only needs to obtain the fresh tobacco leaf image, extract the leaf feature parameters, and input them into the fresh tobacco leaf part discrimination model to directly obtain the part discrimination result, solving the problem of low efficiency caused by manual discrimination of parts before baking and then inputting, which is beneficial to the digital customization of the baking process.

[0029] (2) For different tobacco varieties, the present invention can establish different fresh tobacco leaf part discrimination models, facilitating the rapid determination of the parts of different varieties of fresh tobacco leaves and the judgment of the baking process.

[0030] (3) For newly harvested fresh tobacco leaves, only need to place the fresh tobacco leaves at the shooting position to automatically collect their images, and automatically output the part of the current fresh tobacco leaf based on machine vision recognition, and then match the corresponding baking process according to the fresh tobacco leaf part, realizing the rapid and accurate judgment of the fresh tobacco leaf baking process and improving work efficiency. Description of the Drawings

[0031] Figure 1 It is a flowchart of a fresh tobacco leaf part recognition method based on machine vision according to an embodiment of the present invention.

[0032] Figure 2 Schematic diagram of the image processing effect of fresh tobacco leaves according to an embodiment of the present invention.

[0033] Figure 3 Schematic diagram of the extraction of leaf characteristic parameters according to an embodiment of the present invention. Detailed implementation manners

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0035] As Figure 1 shown, the present invention provides a method for identifying the parts of fresh tobacco based on machine vision, including steps of collecting fresh tobacco leaf images, processing fresh tobacco leaf images, and discriminating the parts of fresh tobacco leaves. Among them, the step of collecting fresh tobacco leaf images is implemented by a camera. The captured image is input into the step of processing fresh tobacco leaf images for image processing to obtain the leaf characteristic parameters of the fresh tobacco leaf image. The leaf characteristic parameters are input into the step of discriminating the parts of fresh tobacco leaves, and the identified parts of fresh tobacco leaves are output, realizing the automatic machine recognition of the parts of fresh tobacco leaves, and solving the problems of low recognition efficiency and low recognition accuracy caused by manual recognition of the parts of fresh tobacco leaves.

[0036] In the step of collecting fresh tobacco leaf images, a leaf collection library of different parts of fresh tobacco leaves is established through harvesting a large number of fresh tobacco leaves. A standard scale is placed beside the fresh tobacco leaves during harvesting and photographing, which is convenient for extracting the pixel data of the standard scale in the preprocessed image.

[0037] As an implementation manner, when photographing fresh tobacco leaf images, the harvested fresh tobacco leaves can be laid flat in the center of a black background cloth, with a standard scale placed beside. The camera is fixed at a preset distance from the ground, and the lens is adjusted to be perpendicular to the ground for image collection.

[0038] After setting up the camera, background cloth and standard scale, for newly harvested fresh tobacco leaves, only need to place the fresh tobacco leaves at the shooting position to automatically collect their images, and automatically output the parts of the current fresh tobacco leaves based on machine vision recognition. Then, the corresponding baking process is matched according to the parts of fresh tobacco leaves, realizing the rapid and accurate determination of the fresh tobacco leaf baking process and improving work efficiency.

[0039] When establishing the leaf collection library of different parts of fresh tobacco leaves, the number of pictures of upper leaves, middle leaves and lower leaves is not less than 100. Such a setting can meet the training and testing requirements of the fresh tobacco leaf part discrimination model, making the constructed model meet the accuracy requirements.

[0040] In the image processing steps of fresh tobacco leaves, it includes image preprocessing, leaf contour extraction, and leaf feature parameter extraction. This process can be implemented within a computing device. After the leaf image captured by the camera is input into the computing device, image preprocessing, contour extraction, and leaf feature parameter extraction are performed in sequence, and the maximum leaf length, maximum leaf width, and leaf width ratio of the leaf are output.

[0041] Image preprocessing aims to separate the leaf from the background through a series of processes on the originally captured leaf image, so as to facilitate the subsequent extraction of the leaf contour curve. Preferably, image preprocessing includes: image grayscale conversion, Gaussian filter smoothing, grayscale image binarization, and morphological processing to achieve the separation of the leaf from the background.

[0042] Furthermore, after preprocessing the pictures in the tobacco leaf harvesting picture library, the canny operator is used to extract the contour edge curve of the flue-cured tobacco for the preprocessed pictures. The effect diagrams of image preprocessing and contour extraction are as Figure 2 shown.

[0043] In the leaf feature parameter extraction step, an external rectangle of the leaf contour is constructed based on the extracted flue-cured tobacco contour edge curve; leaf feature parameters are extracted based on the external rectangle of the leaf contour. Since the leaf is photographed with a standard scale as a reference, in the obtained image, the maximum leaf length, maximum leaf width, and leaf width ratio can be calculated according to the pixel proportion of the standard scale.

[0044] As Figure 3 shown, the number of pixel points of the line connecting the two points with the maximum distance on the leaf contour is used as the maximum leaf length (such as the number of pixel points between T and B), and the number of pixel points of the line connecting the two points that are perpendicular to the leaf length and have the maximum distance on the leaf contour is used as the maximum leaf width (such as the number of pixel points between L and R), and the leaf width ratio is calculated based on the maximum leaf length and the maximum leaf width.

[0045] In the fresh tobacco leaf position discrimination step, according to the maximum leaf length and the length-width ratio, the fresh tobacco leaf position discrimination model is established as: position = a + b × maximum leaf length + c × maximum leaf width + d × length-width ratio.

[0046] Before constructing the fresh tobacco leaf position discrimination model, first obtain the leaf samples for model construction, and then professional personnel who have been engaged in flue-cured tobacco production work for a long time conduct position discrimination and annotation on the leaf samples, that is, label the upper leaves, middle leaves, and lower leaves in the tobacco leaf harvesting picture library respectively, and randomly select the training set and the test set at a preset ratio to obtain the leaves with position annotation for model training and testing.

[0047] Based on the training set, the parameters a, b, c, and d of the fresh tobacco leaf position discrimination model are calculated using the stepwise regression equation, and the accuracy is tested using the test set. Finally, a fresh tobacco leaf position discrimination model that meets the accuracy requirements is obtained.

[0048] Preferably, the position discrimination model of fresh tobacco leaves of different varieties can be calculated according to the stepwise regression equation. By obtaining a large amount of data on the maximum leaf length and leaf width ratio of fresh tobacco leaves of the same variety, as well as the position annotation information of these fresh tobacco leaves, the position discrimination model of fresh tobacco leaves of this variety is calculated using the stepwise regression equation.

[0049] When performing the stepwise regression equation calculation, the variance inflation coefficient of each leaf feature parameter is statistically analyzed, and the leaf feature parameters with a variance inflation coefficient greater than the preset value are excluded. For fresh tobacco leaves of different varieties, there may be different and insignificantly high feature parameters during the stepwise regression equation calculation. Excluding this feature parameter may result in different varieties of fresh tobacco leaf position discrimination models containing different feature parameters. For example, it may only include the maximum leaf length and leaf width ratio, or only include the maximum leaf width and leaf width ratio, or include both the maximum leaf length, maximum leaf width, and leaf width ratio.

[0050] In practical applications, only the leaf feature parameters included after the stepwise regression calculation need to be input into the fresh tobacco leaf position discrimination model, and the position of the current leaf can be output, and the corresponding baking process can be matched according to the current position.

[0051] The output end of the fresh tobacco leaf position discrimination model can be connected to the baking controller, and the baking controller automatically matches the corresponding baking process according to the position of the fresh tobacco leaf, realizing an intelligent operation process for determining the position of the fresh tobacco leaf and judging the baking process.

[0052] Embodiment

[0053] Taking Yunyan 87 as an example, tobacco fields with standardized management, representative tobacco plants are selected and harvested during the mature period. The harvesting position of the upper leaves is at the 15th to 18th leaf positions, the harvesting position of the middle leaves is at the 9th to 12th leaf positions, and the harvesting position of the lower leaves is at the 5th to 7th leaf positions.

[0054] Three professionals who have been engaged in flue-cured tobacco production for a long time are invited to judge the position of fresh tobacco leaves. A total of 480 fresh tobacco leaf samples are collected, including 148 lower leaves, 168 middle leaves, and 164 upper leaves. The harvested fresh tobacco leaves are laid flat in the center of a black background cloth, and the Olympus XZ-1 CCD camera is fixed on a tripod, about 1.5 m above the ground. The lens is adjusted perpendicular to the ground using a level to collect images.

[0055] The original image of fresh tobacco leaves is a 24-bit true color image with a resolution of 3648×2736 pixels. The upper leaves, middle leaves, and lower leaves are labeled as 1, 2, and 3 respectively, and the training set and test set are randomly selected at a ratio of 4:1.

[0056] The original image of fresh tobacco leaves is converted into a grayscale image. Image preprocessing such as separating the leaves from the background in the image is achieved through Gaussian filtering smoothing, grayscale image binarization, and morphological processing. The Canny operator is used to extract the contour edge curve of the fresh tobacco leaves, and then the leaf feature parameters of the fresh tobacco leaves are extracted.

[0057] The number of pixel points on the line connecting the two points with the largest distance on the leaf contour is used as the maximum leaf length, and the number of pixel points on the line connecting the two points that are perpendicular to the leaf length and have the farthest distance on the leaf contour is used as the maximum leaf width. The leaf width ratio is calculated based on the maximum leaf length, maximum leaf width, and maximum leaf width.

[0058] Based on the maximum leaf length, maximum leaf width, and leaf width ratio, combined with the stepwise regression equation, the discriminant model for the position of fresh tobacco leaves is calculated as position = -5.047 + 0.051×maximum leaf length + 1.265×length-width ratio. The maximum leaf width with low significance is removed from the final model.

[0059] In the above model, it is stipulated that y = 1 represents the lower leaves, y = 2 represents the middle leaves, and y = 3 represents the upper leaves. Whichever position the calculated position value is close to, it is confirmed to belong to that position. For example, if the calculated position value is less than 1.5, it is the lower leaves; if it is between 1.6 and 2.5, it is the middle leaves; if it is greater than 2.5, it is the upper leaves.

[0060] Based on the above model, the position of fresh tobacco leaves of Yunyan 87 is discriminated, and the discrimination results are shown in Table 1.

[0061]

[0062] Table 1 Recognition results of the positions of fresh tobacco leaves of Yunyan 87

[0063] In the description of this specification, the descriptions referring to terms such as "one embodiment", "certain embodiments", "schematic embodiments", "examples", "specific examples", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying the parts of fresh tobacco leaves based on machine vision, characterized in that, The method comprises: Establish a leaf harvesting gallery of different parts of fresh tobacco leaves, and place a standard ruler next to the fresh tobacco leaves when taking photos of them; mark the upper leaves, middle leaves, and lower leaves in the leaf harvesting gallery, respectively, and randomly select training sets and test sets at a preset ratio to train and test the constructed fresh tobacco leaf part discrimination model; Preprocess the images in the leaf harvesting image library, and use the Canny operator to extract the edge curve of the flue-cured tobacco contour from the preprocessed images; The circumscribed rectangle of the leaf contour is constructed based on the extracted edge curve of the flue-cured tobacco contour; The leaf feature parameters are extracted based on the circumscribed rectangle of the leaf contour. The number of pixels on the line between the two points with the largest distance on the leaf contour is taken as the maximum leaf length. The number of pixels on the line between the two points with the longest distance perpendicular to the leaf length on the leaf contour is taken as the maximum leaf width. The leaf width ratio is calculated based on the maximum leaf length and the maximum leaf width. According to the maximum leaf length, maximum leaf width and length-to-width ratio, the fresh tobacco leaf part discrimination model was established as follows: part = a + b × maximum leaf length + c × maximum leaf width + d × length-to-width ratio; Based on the training set, the parameters a, b, c, and d of the fresh tobacco leaf part discrimination model were calculated using the stepwise regression equation, and the accuracy test was performed using the test set. Finally, a fresh tobacco leaf part discrimination model that met the accuracy requirements was obtained. The position discrimination model of different varieties of fresh tobacco leaves was calculated based on the stepwise regression equation: by obtaining a large amount of maximum leaf length and leaf width ratio data of the same variety of fresh tobacco leaves, as well as the position annotation information of these fresh tobacco leaves, the position discrimination model of the variety of fresh tobacco leaves was calculated using the stepwise regression equation.

2. The method for identifying the position of fresh tobacco leaves based on machine vision according to claim 1, wherein The method further comprises: inputting the maximum leaf length and leaf width ratio of the fresh tobacco leaf into a fresh tobacco leaf position discrimination model, outputting the position of the current leaf, and matching the corresponding baking process according to the current position.

3. The method for identifying the fresh tobacco leaf position based on machine vision according to claim 1, wherein, The position discrimination model of different varieties of fresh tobacco leaves was obtained based on the stepwise regression equation.

4. The method for identifying the fresh tobacco leaf position based on machine vision according to claim 3, characterized in that When performing stepwise regression equation calculation, the variance expansion coefficient of each blade characteristic parameter is counted, and the blade characteristic parameters with variance expansion coefficients greater than a preset value are eliminated.

5. The method for identifying the position of fresh tobacco leaves based on machine vision according to claim 1, wherein When taking images of fresh tobacco leaves, lay the harvested fresh tobacco leaves flat in the center of a black background cloth, place a standard ruler next to it, fix the camera at a preset distance from the ground, and adjust the lens to be perpendicular to the ground to capture the image.

6. The method for identifying the positions of fresh tobacco leaves based on machine vision according to claim 1, wherein, When establishing a library of images of fresh tobacco leaves harvested from different parts of the leaves, the number of images of upper leaves, middle leaves, and lower leaves should be no less than 100.

7. The method for identifying the parts of fresh tobacco leaves based on machine vision according to claim 1, wherein, The preprocessing of the collected images further includes: image grayscale, Gaussian filter smoothing, grayscale image binarization and morphological processing to achieve separation of leaves from background.

Images