Flue-cured tobacco aroma type and quality judgment intelligent evaluation method

An evaluation method and flavor type technology, applied in character and pattern recognition, computer material science, instruments, etc., can solve problems that are susceptible to environmental and personal factors, hazards, tobacco leaf and cigarette style characteristics, etc., to reduce detection costs and avoid damage effect

Inactive Publication Date: 2017-03-22

FUJIAN AGRI & FORESTRY UNIV

3 Cites 6 Cited by

AI-Extracted Technical Summary

Problems solved by technology

At present, the domestic cigarette industry enterprises have not yet formed a unified evaluation standard for the evaluation of the quality and style of tobacco leaves, and the evaluation standards for tobacco leaves are mostly affected by the style and characteristics of their own cigarettes

[0005] To sum up, the current method of evaluating the quality and aroma characteristics of tobacco leave...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Abstract

The invention relates to a flue-cured tobacco aroma type and quality judgment intelligent evaluation method. The method is characterized by screening out aroma classification and aroma score principal component factors from tobacco aroma substances through a GRRF method; establishing a tobacco aroma style judgment model through GRRF+SVM and GRRF+RF methods; analyzing evaluation performance of the two methods through a 10 fold cross-validation method, wherein for classification evaluation results, adopting accuracy ACC and receiver operating characteristic curve as evaluation criterion, and for regression evaluation results, adopting Pearson's correlation coefficient and mean square error as evaluation criterion, and determining an optimal plan of a tobacco aroma type judgment model; and on the basis of the above, constructing a tobacco aroma type judgment online tool and carrying out evaluation on the aroma type and quality of tobaccos to be detected. The method can carry out analysis and judgment on aroma type characteristics of the tobaccos in different ecological areas objectively and accurately.

Application Domain

Character and pattern recognitionChemical machine learning +4

Technology Topic

FlueDiscriminative model +6

Image

Examples

Experimental program(1)

Example Embodiment

[0050] The technical solution of the present invention will be described in detail below in conjunction with the drawings.

[0051] The present invention provides an intelligent evaluation method for determining the flavor and quality of flue-cured tobacco, which is implemented according to the following steps:

[0052] Step S1: Using GRRF (Guided Regularized Random Forest) method to screen the principal component factors of tobacco aroma classification and aroma score, 22 aroma classifications and 19 aromas were selected from 92 aroma substances Principal component factors of aroma score;

[0053] Step S2: use GRRF+SVM (support vector machine) and GRRF+RF (random forest) to establish a tobacco leaf aroma style discrimination model;

[0054] Step S3: Use the 10 fold cross-validation method to analyze the evaluation performance of the two methods. For the classification evaluation results, the accuracy rate ACC (Accuracy) and the receiver operating characteristic curve (ROC) are used as Evaluation criteria: For the regression evaluation results, Pearson Correlation Coefficient (PCC) and Mean Square Error (MSE) are used as evaluation criteria to determine the tobacco aroma style discriminant model; in this example, it is determined to use GRRF+SVM to establish tobacco aroma Style discrimination model;

[0055] Step S4: Use the PHP and R platform to construct an online tool for judging the flavor and style of tobacco leaves to evaluate the flavor and quality of the tobacco to be tested.

[0056] Further, in this embodiment, the process of using the GRRF method to screen the principal component factors of tobacco aroma classification and aroma scores further includes the following steps:

[0057] Step S11: first construct a random forest model on the sample set before using GRRF.

[0058] Step S12: After the construction of the random forest is completed, GRRF will traverse all the nodes of the original random forest again and calculate the variable importance. For each variable X i The models all give corresponding importance scores, which are calculated according to the following formula:

[0059]

[0060] Where k is the number of decision trees in the forest, x i Is the i-th variable, v is the use of x in the tree i Split node, Gain(X i ,V) is the decrease in Gini impurity of the node.

[0061] Further, for node v, from node v to its left and right child nodes, the reduction in Gini impurity is:

[0062] Gain(x i ,v)=Gini(x i ,v)-(w R Gini(x i ,v R )+w L Gini(x i ,v L ));

[0063] Among them, R and L respectively represent the left and right child nodes of v, W R And W L It is the weight value of the Gini impurity reduction of the left and right child nodes.

[0064] For node v, the Gini impurity is calculated as follows:

[0065]

[0066] Where p c Is the probability of category c in node v, and is the number of samples in category c divided by the total number of samples in node v.

[0067] Step S13: For Gain(X i ,V), but also to be processed as follows:

[0068]

[0069] Among them, F represents the set of features that have been selected when traversing the nodes. When a feature is selected, a certain degree of punishment is required. The size of the punishment is determined by the coefficient λ. The larger the λ, the smaller the punishment. λ is calculated as follows:

[0070] λ=1-γ(1-imp’);

[0071] Among them, γ is the only parameter that GRRF needs to determine. According to the experimental results, when γ=0.1, the selected features enable the model to maintain a good accuracy rate. Where imp’ is the importance score given by the original random forest.

[0072] Step S14: In the end, GRRF will also give the importance score of each variable based on the above principle, and those variables with variable importance scores greater than zero are the principal components selected by GRRF.

[0073] Further, in this embodiment, an RF (Random Forest) classifier is used to construct a fragrance style discrimination model based on GRRF. In this process, three main parameters need to be set: the number of decision trees in the forest ntree, and the random internal nodes Select the number of attributes mtry and the minimum sample number nodesize of the end node. The basic idea is:

[0074] One is to use bootstrap sampling technology to extract ntree training sets from the original data set;

[0075] The second is to build a classification regression tree for each bootstrap training set, and a total of ntree decision trees are generated to form a "forest", and these decision trees are not pruned (unpruned). During the growth of each tree, instead of selecting the optimal attribute among all M attributes as internal nodes for branching (split), but selecting the optimal attribute from randomly selected mtry≤M attributes to branch;

[0076] The third is to aggregate the prediction results of ntree decision trees, and use voting to determine the category of the new sample.

[0077] Further, in this embodiment, the MVB (Support Vector Machine) classifier is used to construct a fragrance style discrimination model based on GRRF. The basic idea is to find an optimal classification hyperplane (separating hyperplane) so that the two classes are closest to each other. The margin between the sample points of is maximized. The sample points on the boundary of the maximization edge are called support vectors, and the middle section of the edge is the optimal classification hyperplane. Points that are misclassified by the edge, their weight will be reduced to reduce their impact. When the data is linearly inseparable, the data points are mapped to the high-dimensional space through the kernel function to make it linearly separable.

[0078] Further, in this embodiment, a tobacco aroma style discrimination model is established based on GRRF+SVM, and an online tool for tobacco aroma style discrimination is constructed using PHP and R platforms. Such as figure 2 As shown, the selection operation items are provided at the top of the main page of the tobacco aroma style evaluation system, including: comprehensive quality evaluation, aroma evaluation, usage guide, function introduction, and data download.

[0079] Such as image 3 As shown, the following takes the sample of DXY010 (Overseas Village, Xinhua Township, Xinping County, Yuxi City) as an example to introduce the use of the online tobacco quality evaluation system and the accuracy of the constructed model evaluation.

[0080] First, select the flavor evaluation menu in the main page of the tobacco aroma style evaluation system, such as Figure 4 Shown. Then, follow the prompts of the pop-up menu input box, enter the test results of the 22 aroma substances that determine the aroma type of the DXY010 sample, and click Submit Data. The result is Figure 5 Shown: The fragrance type of the DXY010 sample obtained is "Fresh fragrance type". After obtaining the results of the aroma style type of the DXY010 tobacco leaf sample, if you need to further obtain the aroma score of the tobacco leaf sample, please click "Continue", and the tobacco aroma score main component content input page will appear, such as Image 6 Shown. Then, follow the prompts of the pop-up menu input box, enter the test results of the 19 aroma substances that determine the aroma score of the DXY010 sample, and click "Submit Data". The result is Figure 7 Shown: The aroma score of DXY010 sample is "4.4".

[0081] In this embodiment, Table 1 is the principal component factors of 22 aroma classifications obtained by screening, and Table 2 is the principal component factors of 19 aroma scores obtained through screening.

[0082] Table 1

[0083] Serial number Substance name Serial number Substance name X1

[0084] Table 2

[0085] Serial number Substance name Serial number Substance name X1

[0086] According to the above results, it can be seen that the intelligent evaluation system for the determination of flavor and quality of flue-cured tobacco has the same effect on the determination of flavor style and flavor score of tobacco leaves. image 3 The results of manual evaluation are basically the same. The result of manual evaluation is 4.5, and the difference between the two is 0.1, which is within the allowable range of error. It shows that the system can objectively and accurately determine the aroma style type and aroma score of tobacco leaves, and can be used as a cigarette raw material formula to determine the aroma style type and aroma score of tobacco raw materials.

[0087] The above are the preferred embodiments of the present invention. Any changes made according to the technical solution of the present invention and the resulting functional effects do not exceed the scope of the technical solution of the present invention belong to the protection scope of the present invention.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to view more

PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine

Login to view more

Classification and recommendation of technical efficacy words

Avoid harm
Reduce testing costs

Flue-cured tobacco aroma type and quality judgment intelligent evaluation method

AI-Extracted Technical Summary

Problems solved by technology

Abstract

Image

Examples

Example Embodiment

PUM

Description & Claims & Application Information

Similar technology patents

Grinding amount on-line monitoring device at ring-polishing stage and grinding amount on-line monitoring method

Method for rapidly testing service life of flexible tube and test bed thereof

Auxiliary identification method of geographical origin of wheat

Method for simultaneously detecting 26 organic acids and inorganic negative ions in juice

Method for measuring carbon content in lithium iron phosphate

Classification and recommendation of technical efficacy words

Robot eye-on-hand system structured light plane parameter calibration device and method

Environment-friendly type alkali-free liquid accelerator for sprayed concrete and preparation method thereof

Mechanical self-adaptive auxiliary docking platform

A dos/ddos attack detection method

System and method for remotely programming a patient medical device

Intelligent toilet

Apparatus for high density low cost automatic test applications

Pavement smoothness detection method based on mobile terminal and intelligent pavement information real-time monitoring system

System and method for network test

Pavement health condition rapid detection system and method