Study On Grading Of Tobacco Leaf Based On Fuzzy Rules

The grading of flue-cured tobacco directly determines the quality of its product. In order to improve the grading accuracy and efficiency, we must have scientific and rational tobacco quality evaluation criteria. At present, tobacco grading mainly rely on people’s senses, this grading method has a strong subjectivity and arbitrariness. Different experts may give different level results for the same tobacco leaf, which will cause many problems and contradictions. Therefore, the intelligent grading of tobacco research is necessary. By extracting the characteristic parameters of tobacco image, auto-grading is studied based on the fuzzy rules. The main studies are as follow:1. Tobacco image acquisition and image pre-processing. By assembling industrial CCD cameras, frame grabbers, computer, daylight lamp and other equipment, tobacco samples images were collected in a closed black box and saved in the computer. Digital image processing is used to obtain the image preprocessing. The median filter is chosen to move the noise of the image. The regional threshold method combined with the small area removing method is used to segment tobacco leaf from the image background. Finally the contour extraction method is used to extract the border information of a tobacco leaf.2. Feature extraction of tobacco leaves. Image processing techniques is adopt to extract the feature. Include shape features: length, width, aspect ratio, perimeter, area, degree of circularity, damaged degree, etc. Color features: the average of red, green, blue, hue, luminance and saturation and their variance. Texture features: correlation, contrast, energy, homogeneity.3. Feature selection and grading of tobacco leaves. First make the leaves into groups, and then grading in the group. Divided into two parts, first property filter module. Get the weight of each property in the classification by the characteristic parameter attribute correlation analysis. Remove the minimum correlation property attributes, and establishing a classification model to compare the accuracy. Second tobacco graded modules. Input the best combination of features into the classifier based on fuzzy rules, and it can get the membership values of each tobacco leaf. Then output the grade of each tobacco leaf according to the maximum degree of membership.Through the feature selection and the correct rate of tobacco classification, the analysis conclusion is that using the classifiers based on fuzzy rules can effectively complete the task of grading of tobacco leaf, and also with a high accuracy rate.