Application Of Tobacco Classification Based On K-nearest Neighbor

At present, tobacco classification is relied on the smell and taste of the professional who has received specialized training. This approach consumes a large amount of manpower and material resources, and it is difficult to guarantee the objectivity and accuracy of tobacco classification, because of artificial grading method has very strong subjectivity and fuzziness. In addition to the above, the approach is increasingly unable to meet the requirement of the quality of tobacco products. In order to overcome the defects of artificial grading method, tobacco’s intellectual classification based on image processing technology combined with k nearest neighbor is adopted to improve accuracy of tobacco classification. Tasks of the paper mainly include the following four parts:1. Image acquiring and preprocessing of tobacco leaf. Select experiment equipment to acquire tobacco leaf in real time, then some preprocessing has been done to the image of tobacco leaf. By the comparison of different image smooth, background segmentation and edge extraction, the median filter is used to smooth the image and eliminate the noise of the leaf image, the method of threshold value is employed to segment tobacco leaf from the background, and the contour extraction method is chosen to extract the edge information of the tobacco leaf.2. Features extracting of tobacco leaf. Image processing technique is adopted to extract seven shape features, twelve color features and four texture features. Shape features include length, width, aspect ratio, area, perimeter, breakage rate, and the degree of circularity. Color features include the average of red, green, blue, hue, saturation, intensity and their variance. Texture features include energy, entropy, contrast and correlation.3. Feature selection. The paper extracts 23 features which may affect the result of tobacco classification, however, these features’ relevance and redundancy increase the complication of tobacco classification and influence the accuracy of tobacco classification. In order to reduce the complexity and improve the accuracy of the tobacco’s intellectual classification, the backward feature selection based on k nearest neighbor is adopted to select the best features which are used in grading of tobacco leaves.4. Tobacco’s intellectual classification. Aiming at overcoming the short of traditional KNN, an improved KNN(KKNN, k nearest neighbor based on K-Means) is proposed. KKNN uses entropy method to give different weight value to different features, at the same time, uses k-means to cluster training data set to decrease the volume of training data, lastly, uses the average Euclidean distance to decide the category. The experimental results show the improved algorithm and backward feature selection can improve tobacco’s classification accuracy.