Tobacco Classification Based On Image Processing

Tobacco is an important economic crop. Its quality is the basis of cigarette products. If we want to improve the quality of tobacco products, we should have scientific and rational leaf quality standards and accurate grading identification. Flue-cured tobacco leaves are divided into 42 grades in the current grading standard. It is based on seven grading factors, including leaf structure, maturity, color, identity, oil, damage and length. The descriptions of classification standard are rather vague.At present, tobacco classification is based on experience in our country. There are no specific quantitative targets. The quality is still determined by the sense of taste and smell of professional who has received special training. It also takes considerable human, material and financial resources. And obviously, the classification is often affected by many factors. In practice, different professionals may give different results for the same leaf. Thus it may cause many problems and contradictions. The current grading method is unable to adapt to modern society requirements of industrialization, modernization, automation and information. In order to overcome these short comings, more and more researchers have begun to study intelligent classification methods.Computer Vision is an important branch of computer science. It involves computer, image processing, pattern recognition, artificial intelligence, signal processing, optics and other fields. So my paper is about tobacco classification based on image processing. I have done the following wok:First, I have gotten different tobacco images. Second, use the digital image processing technology to preprocess these images and get their digital image features. In this study I extracted shape features, color features and texture parameters. Shape features include perimeter, area, roundness factor, length, width, aspect ratio. Leaf texture parameters include energy ENT, angular second moment ASM, contrast CON, relevance of COR, consistency of U, the average gray level, gray level variance, etc.Then through comparison and analysis, I select artificial neural network BP algorithm back propagation to build the mathematical model of tobacco classification. This method has advantages over other ones at precision and efficiency. At the end of my paper, experiment results have been given to demonstrate the effectiveness and practicality. Experiment has shown that the training samples are obtained 100% recognition rate. All the test samples were over 73.6%on average, the highest recognition rate was 82.3%. Compared to other classification based on chemical composition and spectrum, this method is non-destructive and lower-cost.It is more practical and objective compared to manual classification.especially in large-scale tobacco purchase. It provides a new way for tobacco classification.Although this study has achieved a good classification results relatively, because of limited research time and condition, the study also has some deficiencies. Further research should be done in the following work.