Study And Realization Of Digital Elements Characterization And Classification Of Tobacco Leaf Characteristics

Traditional tobacco classification is implemented according to the national standards of flue-cured tobacco,and the process is often carried out by classifying workers through their senses and combined with experience,which leads to strong subjectivity of classifying individuals and easy to occur grade misclassification.The description of tobacco grade classification in the current national standards for flue-cured tobacco is not quantitative analysis,which is not conducive to the objective classification of tobacco,and does not supplement and explain the diversification of tobacco varieties and tobacco producing areas.In view of the above problems,the shape and color features of tobacco leaves were extracted based on machine vision and converted into digital features.Then,rules for tobacco parts division and grade division were extracted based on CART decision tree algorithm to realize the purpose of quantifying national standards for flue-cured tobacco.The specific research content and work are as follows:(1)In view of the excessive subjectivity and grade misclassification caused by manual classification of tobacco leaves based on experience and sense,a method based on machine vision was proposed for image processing of collected tobacco leaves,and then the characteristics of tobacco leaves were represented by digital elements to extract the shape characteristics and color characteristics of tobacco leaves.(2)Aiming at the standard quantification of flue-cured tobacco,taking commonly grade tobacco as an example,3,174 tobacco leaf data of commonly used tobacco leaf were obtained by cleaning the extracted tobacco leaf feature data set.Firstly,characteristic correlation analysis was conducted on the classification of tobacco leaf for each element in the tobacco leaf feature data set.The position or grade of tobacco leaves in this data set was taken as the category of tobacco leaves,and the classification of CART decision tree was made according to the importance degree of characteristics and the superimposed characteristic factors with strong correlation.Based on the prior knowledge and the manual classification process of tobacco leaves,tobacco leaves were first divided into positions,then tobacco grades.After the construction of tobacco parts classification feature rules,then construction of upper,middle and lower parts of tobacco leaves classification feature rules.The results showed that the accuracy of tobacco parts classification was 76.85%.The accuracy of tobacco grade classification was only 68.52% when tobacco parts were not divided.The classification accuracy of upper tobacco,middle tobacco and lower tobacco was 87.18%,83.26% and 92.92%,respectively.(3)A real-time tobacco classifying software is designed based on C#.Through this software,tobacco classifying workers can not only view the shape and color characteristics of tobacco leaves in real-time detection,but also learn the image of tobacco leaves after pretreatment.