Method For Automatic Grading Of Tobacco Based On Machine Vision

Tobacco grading is the first step for tobacco leaves from agricultural products to commodities, which is a key link to determine the quality of tobacco. The traditional grading method is manual grading. Namely, some staff who had been trained professionally judge tobacco grades with the eyes, the senses of touch and the theory knowledge they had mastered about tobacco grading and the grading experience they had accumulated. However manual grading is strong subjective and unstable. More objective and stable methods need to be researched for tobacco grading, and more automatic systems need to be developed, which has important practical significance.The main work includes these following points:(1). Set up the system of image collection. In order to ensure the light intensity of consistent and uniform for every tobacco leaf image, an airtight image collection system must be established. The system includes digital camera, lamp, power supply, loading platform, and closed box.(2). Preconditioning methods of tobacco images. Image preconditioning mainly includes filtering, binary, segmentation. Median filter can protect the details of tobacco images and filter noise. In order to extract the geometric characteristics of tobacco leaves, tobacco images which had filtered should be binary with the iterative threshold method. In order to extract the color feature of tobacco leaves, the filtered images of tobacco leaves need to be segmented based on Region-Growing method. Reject the background, folds, and veins of tobacco images, and obtain the target objects.(3). Extraction methods of the appearance characteristics for tobacco leaves.Appearance characteristics of tobacco leaves include shape and color characteristics.The length, the width, the ratio of length and width, the area and the tip rectangle degree need to be extracted with the Minimum Enclosed Rectangle. The tobacco leaves' color characteristics of the Hue, Saturation, Channel A and Channel B need to be extracted in the color space model of HSI and LAB.(4). Algorithms of tobacco automatic grading based on fuzzy pattern recognition.First of all, it is the universe with the space of tobacco grades and fuzzy pattern with themean of each appearance grade characteristics vector. The membership of tobacco grades space for various appearance characteristics should be calculated based on the trapezoidal and semi trapezoidal membership function. Then the membership of various appearance characteristics should be weighted and averaged. At last, the tobacco grades would be determined by maximum membership degree law.(5). A tobacco automatic grading software system was developed based on MATLAB platform, and the software system include image preconditioning, extraction of appearance characteristics and identification of tobacco leaf grades.(6). Case analysis. There are 234 tobacco samples(total 22 grades). 148 samples were used as the modeling set and the reminder samples(total 86 samples) formed the prediction set. With tobacco leaves' appearance characteristics of modeling set, tobacco automatic grading model were established based on Fuzzy Pattern Recognition, and the grouping and grading experiment were done for prediction set. The correct rate of grouping and grading were 93.02% and 80.23%. It shows that the method of tobacco automatic grading based on machine vision had reached or even exceeded the level of manual grading in the shape and color of tobacco leaves.