An online identification method of tobacco leaf maturity based on image processing technology

Abstract

The invention discloses an online discrimination method for tobacco leaf maturity based on image processing technology, comprising the following steps: 1) collecting images of tobacco leaves of different maturity in the lower / middle / upper parts of mature tobacco plants; 2) preprocessing the collected images online to filter and reduce noise and extract edges and contours; 3) extracting eigenvalues ​​of the preprocessed images online; 4) using the mean and standard deviation of these eigenvalues ​​as inputs of a support vector machine to establish a network-side multi-classification model, and finally determining the maturity level of tobacco leaves. The method of the invention can realize online identification of tobacco leaf maturity before picking, does not require (carry) expensive equipment, only needs to use a mobile phone or camera to collect images of tobacco leaves, and can analyze and determine the maturity of the images online, and does not require high professional ability of users; the invention ensures accurate prediction of the maturity results of each leaf before harvesting, improves efficiency, and saves labor resources.

Description

Technical Field

[0001] The invention belongs to the technical field of tobacco leaf appearance quality evaluation, and specifically relates to an online tobacco leaf maturity determination method based on image processing technology. Background Art

[0002] Tobacco leaf maturity refers to the degree of maturity of tobacco leaves, which includes two aspects: one is the degree to which tobacco leaves grow and develop under sufficient nutritional conditions, that is, the maturity of fresh leaves in the field. The second is the degree to which mature tobacco leaves are harvested and matured after being conditioned. When tobacco leaves are mature, their appearance characteristics and physiological characteristics show different characteristics, which are the criteria for judging whether the tobacco leaves are mature or not. They are usually divided into under-mature, falsely mature, moderately mature, fully mature, and over-mature. At present, the methods used in the harvesting of mature tobacco leaves at home and abroad are not the same. Japan uses the colorimetric card colorimetric method; Zimbabwe judges the maturity of flue-cured tobacco based on the color of color pictures when the tobacco leaves are mature, the quantitative indicators of the flue-curing room test and the drawer test; the United States judges whether the tobacco leaves are mature by picking tobacco leaf samples one week in advance for chemical composition analysis; some people have also proposed to use the conductivity diagnosis method of flue-cured tobacco leaves to judge the maturity of tobacco leaves. In domestic flue-cured tobacco production, the main method used is to combine the appearance characteristics of leaves with leaf age. For example, the maturity of flue-cured tobacco is judged based on the appearance characteristics of tobacco leaves of different maturity, the angle between stems and leaves, and the age of tobacco leaves harvested at the right maturity. However, the traditional visual inspection (touch) method is too general and abstract. There are empirical, subjective and ambiguous problems in its application. There is no unified judgment standard, and it is difficult to master in actual operation. In recent years, with the development of optoelectronic technology, colorimeter, spectrometer and other equipment have been used to collect tobacco leaf maturity data. The application of spectral technology combined with chemometric methods to quantify the maturity of tobacco leaves has been continuously reported. However, this operation constantly requires (carrying) expensive equipment, greatly increases the professional ability requirements of users, and lacks timeliness. At the same time, it will cause certain damage to tobacco leaves during detection, which makes this type of technology unable to be widely promoted.

[0003] There is no fast, convenient and practical technical method for tobacco leaf maturity detection in the field at home and abroad. Therefore, it is extremely important to study a fast and convenient online method for tobacco leaf maturity identification. Summary of the invention

[0004] The first object of the present invention is to provide an online method for distinguishing tobacco leaf maturity based on image processing technology, comprising the following steps:

[0005] 1) Collect images of tobacco leaves of different maturity at the bottom / middle / upper part of mature tobacco plants and upload them to the network database;

[0006] 2) Preprocessing the collected images online to filter, reduce noise, and extract edges and contours of the images;

[0007] 3) Extracting feature values ​​of the preprocessed image online;

[0008] 4) The mean and standard deviation of these eigenvalues ​​are used as inputs of the support vector machine to establish a network-side multi-classification model and ultimately determine the maturity grade of tobacco leaves.

[0009] Beneficial effects of the present invention:

[0010] 1) According to the appearance and shape of tobacco leaves of different varieties, the present invention removes noise, extracts edges, and subtracts background from tobacco leaf images, combines tobacco leaf feature extraction (mainly based on color features) results, and uses the established multi-classification model to quickly identify tobacco leaf maturity results. The entire recognition process uses old methods in a new way and cleverly combines multiple methods of image preprocessing to achieve online identification of tobacco leaf maturity, greatly improving recognition efficiency and accuracy. The method of the present invention includes tobacco leaf image preprocessing (one-dimensional wavelet transform + adaptive fuzzy clustering), eigenvalue extraction (gray level co-occurrence matrix + RGB range limitation), and network-side multi-classification model establishment (support vector machine), which are cleverly applied to the three tobacco varieties studied by the present invention. The result is high recognition accuracy, and each algorithm is independent and complementary to each other. Except for the SVM method, which requires input parameters, other algorithms do not require parameter input, which greatly improves the objectivity of the analysis results. The parameters of the SVM method are optimized through a large number of samples, and can be appropriately adjusted within the deviation range of ±5% of the corresponding parameters, thereby improving the accuracy of the analysis results.

[0011] 2) The method of the present invention can realize online identification of the maturity of tobacco leaves before picking, without the need for (carrying) expensive equipment. It only needs to use a mobile phone or a camera to collect images of tobacco leaves, and the maturity analysis and judgment of the images can be performed online, which does not require high professional ability of users; the present invention ensures that the maturity results of each leaf are accurately predicted before harvesting, rather than quality physical and chemical analysis after picking, which not only improves efficiency and saves labor resources, but also overcomes the disadvantage of unclear maturity judgment in manual harvesting;

[0012] 3) The online determination method of tobacco leaf maturity of the present invention continues the tobacco leaf "quality improvement and efficiency enhancement" project in tobacco leaf production and acquisition work, ensuring the healthy, stable and sustainable development of the flue-cured tobacco industry. At the same time, it plays a good demonstration role in the control of grade quality in acquisition work, solves the subjectivity of relying on the experience of graders and manual grading of tobacco leaves, vigorously eliminates the problem of inconsistent grading vision between industrial enterprises and commercial enterprises, and creates a good environment for tobacco leaf harvesting.

[0013] 4) The online identification method of tobacco leaf maturity of the present invention breaks through the limitation of region and variety, and the personalized classification model efficiently realizes the random sampling and harvesting of tobacco leaf maturity in the whole field. The interests of tobacco farmers are always taken into consideration, and "favor tobacco and face tobacco" are eliminated, so as to purchase tobacco leaves fairly and impartially and stabilize the purchase grade of tobacco leaves. DETAILED DESCRIPTION

[0014] The present invention is further described below in conjunction with the embodiments, but the present invention is not limited in any way. Any changes or substitutions made based on the teachings of the present invention belong to the protection scope of the present invention.

[0015] The present invention discloses an online determination method of tobacco leaf maturity based on image processing technology, comprising the following steps:

[0016] 1) Collect images of tobacco leaves of different maturity at the bottom / middle / upper part of mature tobacco plants and upload them to the network database;

[0017] 2) Preprocessing the collected images online to filter, reduce noise, and extract edges and contours of the images;

[0018] 3) Extracting feature values ​​of the preprocessed image online;

[0019] 4) Use the mean and standard deviation of these eigenvalues ​​as the input of the support vector machine to establish a network-side multi-classification model and ultimately determine the maturity level of tobacco leaves. Modeling with the mean and standard deviation of eigenvalues ​​can effectively eliminate the impact of abnormal tobacco leaves on modeling.

[0020] The image preprocessing adopts wavelet transform method to filter and reduce noise of tobacco leaf image, and adopts adaptive fuzzy C-means clustering algorithm to extract tobacco leaf edge and contour.

[0021] In step 3, a gray level co-occurrence matrix is ​​used to extract eigenvalues.

[0022] The characteristic values ​​are RGB, HSV and texture features of the tobacco leaf image, and the texture features include texture inertia, texture energy, texture entropy and texture correlation of the texture.

[0023] When using the gray-level co-occurrence matrix for modeling and calculation, the actual characteristics of tobacco leaf maturity (dark green→light green→yellow-green→light yellow→dark yellow) were taken into consideration. During the modeling and calculation, the RGB value range was limited to [0 / 100 / 0→···→128 / 128 / 0], and the extracted outliers were eliminated to make the model prediction results more robust.

[0024] The calculation formula of inertia (CON) is as follows:

[0025]

[0026] The calculation formula of energy (ASM) is as follows:

[0027]

[0028] The calculation formula of entropy (ENT) is as follows:

[0029]

[0030] The correlation is calculated as follows:

[0031]

[0032] The tobacco leaves are disease-free tobacco leaves.

[0033] The image pixels of the collected tobacco leaves shall not be less than 600*800.

[0034] The parameters adopted by the decision function of the support vector machine are (c_svc, linear, rho 17.363, -39.238, -68.126, -83.344, -191.902, 34.962, probA-1.06, -0.865, -0.840, -0.752, -0.974, -0.953; probB-0.096, -0.366, -0.218, -0.0278, 0.003, -0.020), and the parameters can be adjusted within a deviation range of ±5%.

[0035] The SVM multi-classification model strategy of the present invention is: first, a nonlinear mapping Ψ is selected to map the n-dimensional sample vector from the original space to the feature space, and the optimal linear decision function is constructed in this high-dimensional feature space. When constructing the optimal decision function, the structural risk minimization principle is utilized, and the concept of interval is introduced at the same time, and the kernel function of the original space is cleverly used to replace the dot product operation of the high-dimensional feature space, avoiding complex operations. In this model, a one-to-one method is selected for classification for the five-classification problem, that is, a decision function is constructed between any two categories. In this case, there are 10 decision functions for the five-classification problem. At the same time, the support vector is the vector relied on when constructing the decision function, and each class has its own support vector, and the number is not necessarily the same. Based on the above situation, the following are listed for each model after modeling: the total number of support vectors, the number of support vectors in each class, the specific support vectors of each class, and the constant term b in the decision function.

[0036] The sampling indirect method of the present invention realizes multi-classification, mainly by combining multiple binary classifiers to realize the construction of multiple classifiers. Common construction methods include "one-to-one" and "one-to-many".

[0037] The "one-to-one" method requires combining n types of training data in pairs to construct n(n-1) / 2 support vector machines. Each support vector machine is trained on two different types of data. The final classification is determined by "voting".

[0038] The "one-to-many" method constructs n support vector machines for n classification problems, and each support vector machine is responsible for distinguishing data from its own class and data from other classes. The classifier constructs a support vector machine for each class, and the kth support vector machine constructs a hyperplane between the kth class and the remaining n-1 classes. The final result is determined by the support vector machine whose output is the largest distance wx+b from the interface.

[0039] During training, samples of a certain category are classified into one category (such as underripe), and the remaining samples are classified into another category (false ripe, ripe, overripe), so that k SVMs are constructed with samples of k categories. During classification, unknown samples are classified into the category with the largest classification function value.

[0040] This study includes four types of samples (that is, 4 labels), they are M1, M2, M3, M4. So when extracting the training set, we extract

[0041] (1) The vector corresponding to M1 is taken as the positive set, and the vectors corresponding to M2, M3, and M4 are taken as the negative set;

[0042] (2) The vector corresponding to M2 is taken as the positive set, and the vectors corresponding to M1, M3, and M4 are taken as the negative set;

[0043] (3) The vector corresponding to M3 is taken as the positive set, and the vectors corresponding to M1, M2, and M4 are taken as the negative set;

[0044] (4) The vector corresponding to M4 is taken as the positive set, and the vectors corresponding to M1, M2, and M3 are taken as the negative set;

[0045] Use these four training sets to train respectively, and then get four training result files. When testing, use these four training result files to test the corresponding test vectors respectively. Finally, each test has a result f1(x), f2(x), f3(x), f4(x). So the final result is the largest of these four values ​​as the classification result.

[0046] The parameter settings and parameter ranges of this support vector machine SVM are as follows: sampling C_SVC type, linear kernel function, param.cache_size = 100; param.eps = 0.00001; param.C = 1.9; param.probability = 1;

[0047] Example 1

[0048] A field experiment on tobacco leaf maturity determination was conducted in Jiuxi Town, Jiangchuan District, Yuxi City. The tobacco variety tested was K326.

[0049] After the tobacco leaves enter the maturity stage, 80 disease-free tobacco leaves of various parts and maturity are randomly harvested, and the harvested tobacco leaves are coded and labeled; the camera function of the mobile phone is turned on, and the tobacco leaves are placed flat on the ground to take pictures. After taking pictures, the basic information of the tobacco leaves (geographic location, variety, part, etc.) is completed, and the pixels of the tobacco leaf images are 600*800 or above; the tobacco leaf images are uploaded to the network database for online analysis and processing: first, the wavelet transform method is used to filter and reduce the noise of the tobacco leaf images, and the adaptive fuzzy C-means clustering algorithm is used to extract the edges and contours of the tobacco leaves. Then, the RGB, HSV, texture inertia, energy, entropy and correlation of the tobacco leaf images are extracted through the gray level co-occurrence matrix, and the mean and variance of the above eigenvalues ​​are used as the final value. These 20 values ​​are input into the support vector machine multi-classification model to determine the maturity of the tobacco leaves.

[0050] The experimental results are shown in Table 1. The results show that compared with the results of the flue-cured tobacco grading experts, the accuracy of the method in distinguishing the maturity of tobacco leaves is as high as 88%.

[0051] Table 1 Test results of tobacco leaf samples in Example 1

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

Claims

1. An online method for distinguishing tobacco leaf maturity based on image processing technology, characterized in that: The following steps are involved: 1) Collect images of tobacco leaves of different maturity at the bottom / middle / upper part of mature tobacco plants and upload them to the network database. The pixels of the collected tobacco leaf images shall not be less than 600*800; 2) Preprocess the collected images online, i.e. filter and reduce noise of tobacco leaf images using wavelet transform method, and extract tobacco leaf edges and contours using adaptive fuzzy C-means clustering algorithm; 3) Extracting eigenvalues ​​of the preprocessed image online using a gray-level co-occurrence matrix, wherein the eigenvalues ​​are RGB, HSV and texture features of the tobacco leaf image, wherein the texture features include texture inertia, energy, entropy and correlation; when using the gray-level co-occurrence matrix for modeling and calculation, the value range of RGB is limited to [0 / 100 / 0···128 / 128 / 0], and the extracted outliers are eliminated; 4) Using the mean and standard deviation of these eigenvalues ​​as input of a support vector machine to establish a network-side multi-classification model, and ultimately determining the maturity grade of tobacco leaves, the decision function of the support vector machine adopts the following parameters (c_svc, linear, rho is 17.363, -39.238, -68.126, -83.344, -191.902 or 34.962, probA is -1.06, -0.865, -0.840, -0.752, -0.974 or -0.953, and probB is -0.096, -0.366, -0.218, -0.0278, 0.003 or -0.020, respectively), and the parameters can be adjusted within a deviation range of ±5%.

2. The method for online identification of tobacco leaf maturity based on image processing technology according to claim 1, characterized in that: The tobacco leaves are disease-free tobacco leaves.

3. The method for online identification of tobacco leaf maturity based on image processing technology according to claim 1 is characterized in that: The method is applied to the identification of the maturity of flue-cured tobacco leaves of K326, Honghua Dajinyuan or Yunyan 87.

Images