Research Of Tobacco Leaves Sorting System Based On Visual Identification

The grading of tobacco before baking have a direct relation with the tobacco baking process, income of the famers and health of smokers. In this paper, computer vision technology was used to grade the maturity and different sizes, focusing on the problem that tobacco grower paid no attention and non-objectivity to grading tobacco in the grading of tobacco before baking.In order to automate in grading process of tobacco, the main research contents and work in this paper as follows:1. In this paper a comeback sorter was proposed using for tobacco sorting, and introduced the structure of the machine. The detailed design of a sorting control system for tobacco leaves was presented in this paper.2. Determine the image pretreatment process of tobacco leaf. In this paper, the median filtering algorithm was used to achieve image denoising, the iterative threshold method was used for image segmentation, and Canny edge operator was used to realized the image edge extraction.3. Identified the tobacco maturity level described by eight characteristics with color features and texture features. Analyzed and compared tobacco leaf characteristics grown in the water land and the dry land. Experiments found out:(1) The scope of color differentiation was the most obvious in the color features of tobacco leaves which are planted in different environment. (2) The difference of contrast and entropy is the maximum in textural features of different maturity tobacco. (3) Tobacco image contrast, homogeneity, correlation and energy of water land were bigger than dry land, while entropy was smaller, and the rate of change of tobacco leaf texture featureis of water land was bigger than the dry land’s.4. Three categorizers were established based on BP neural network algorithm, ELM algorithm and R_ELM algorithm. Experiments showed that the recognition rate of these three methods were all above 80%, but computation speed of extreme learning machine and regular extreme learning machine were 300 times as much as were the BP neural network algorithm, moreover. The recognition rate of classifier with regular extreme learning machine was the highest.4. According to the existing technology, this paper analyzed communication solution of automatic classificationand sorting of tobacco leaf.