Visual Inspection Technology And Application Research Of Tobacco Leaf Surface

At present, as tobacco classification is still by the artificial method that mainly based on the human sensory experience which was subjectivity, uncertainly and instability in China. Basing on color, length, maturity, mutilated, oil content, odor, motley level and leaf structure, the grader divided tobacco leaves into national 42 division. As the method cannot achieve a specific quantitative indicators, it's adverse to the purchase and quality control of tobacco leaf.As computer vision technology development with a high speed, especially the application of machine vision inspection technology which has characteristics of fast speed, high reliability, no injury detection,and etc. In this paper, research of tobacco leaf auto-grading based on machine vision technology was carried from the following several aspects:(1)Research of image acquisition technology of tobacco leafAccording to the characteristics of tobacco leaf and technology demand of tobacco image acquisition, an image acquisition equipment which fit the demand of easy to operate, high illumination uniformity and good imaging effect of tobacco leaf was designed.(2) Research of image preprocessing technology of tobacco leafAs the features used in the process of inspection and classification was more, different preprocessing method should be chose for different abstraction method of features. In this paper, image preprocessing mainly include image filtering, image segmentation and the between-cluster variance binarization method. Image preprocessing laid a good foundation for the image feature extraction of tobacco leaf.(3)Research of characteristics extraction method based on the chrominance spaceExtracting various of color feature of tobacco leaf within RGB and HSV color space model, and extract length, width, aspect ratio and breakage rate, a total of 10 features by using tobacco leaf structure. Minimum distance classification method was used to firstly divided into three parts, then used the color feature divided into groups, and finally hierarchical classification by using characteristic vector set.(4)Research of tobacco classification algorithmThrough experiment verified the method that using features of color, shape, damage rate and etc to built the characteristic vector set for tobacco classification, basing on the minimum distance classification method to developed the model of tobacco leaf auto-classification is feasible.(5)Development of tobacco leaf classification systemBasing on the platform of Visual Studio 2012, using machine vision function library within opencv2.4.4, a tobacco leaf auto-classification program was developed by C++ language. Through combing with the hardware structure of system, a tobacco leaf classification system was developed.