Research On Visual Information Representation And Classification Algorithm Of Tobacco Leaf Feature

In the process of achieving the usability of tobacco products, tobacco grade judgment is as the role of links and bridges, For giving quick and accurate comprehensive evaluation of tobacco grade,it is a key factors that improving quality of tobacco products play. This paper studied expressions and grade algorithms to achieve visual information of tobacco leaves features mainly. The research is carried out from the following aspects mainly:( 1) research of tobacco leaves image acquisition technologiesDesigning image acquisition device according to specific requirements of tobacco leaves image acquisition devices,the device has a series of advantages such as high imaging quality, good illumination uniformity, fixed shooting environment, simple operation and easy realization of engineering.( 2)research of tobacco image preprocessing and segmentationTobacco leaves image processing methods are involved in image enhancement, sharpening and segmentation of image in this paper. Through analysing the experiment, enhancement, sharpening and segmentation of tobacco leaves image respectively selected median filtering, LOG algorithms filtering and Otsu(Otsu algorithm) to achieve. Image by processing segmentation can reflect the real visual information of tobacco leaves. These visual information are a good foundation for expression of the subsequent tobacco feature visual information.(3)research of expression algorithms of tobacco feature visual informationThe mean value of each color component of tobacco is etracted in the RGB color space models, 15 feature indexes such as the mean value of each color component, consistency, smoothness a of tobacco are etracted in the HSV color space models.The R, G, B component and the corresponding mean are used in the RGB distance color formulas to calculate color difference threshold of standard tobacco grades. Quantitative studies of tobacco grading factors used absolute correlation analysis methods of gray system theory and quantitative values of appearance indexes such as maturity and oil characteristics in the extracted and classified factors.According to the central tobacco recognition experimental results, C2 F has highest recognition rate of 72%. C3 L has lowest recognition rate of 64%. The average recognition rate of central tobacco reaches 68%, meeting requirements for the recognition rate during tobacco acquisition process.(4)research of expression algorithms implemention of tobacco leaves feature visual informationBased on the platform OpenCv2.4.4 computer vision library, the color difference algorithm that the paper gived is transplanted by OpenCV, and automatic classification algorithm program is wrighted based on the tobacco features visual information, mathematical models of automatic classification of tobacco is established as well.