Tobacco Stem Based On Machine Vision Detection And Tobacco Leaf Type Identification Method Research

In tobacco production industry, many production procedures are still finished manually, which leads to people have to pay much more time, hard work, and money. Tobacco stem detection and tobacco leaf classification are two classical procedures. One is aiming at removing the stem and the other is aiming at classifying the tobacco leaf to more special leafs. Machine vision based tobacco stem detection and tobacco leaf classification is to achieve tobacco stem automatic detection and automatic type identification of tobacco leaf by using computer vision, machine learning and pattern recognition methods. Researches on the method of dealing with the two problems and devote to provide any research achievements to engineering application are importance and meaningful since it can improve efficiency and cut more production cost. This thesis aims at dealing with the two problems by machine vision. The major work of this thesis includes:Firstly, we validate the feasibility of using infrared hot radiation difference for tobacco stem detection. Based on infrared tobacco image features, we introduce the thought of visual saliency to tobacco infarction stem segmentation, and propose a method based on visual saliency histogram for infrared tobacco infarction stem segmentation, and verify the effectiveness of the algorithm by experiments. Moreover, we calculate the leaf removal criterion parameters correctly base on analyzing the segmentation image.Secondly, we deeply investigate color and texture based feature extraction method, propose a two-fold color and texture fusion based feature extraction method. We apply this feature to sparse representation framework and design a sparse representation based tobacco leaf classification approach. Furthermore, we propose a tobacco leaf classification method based on two-value color vector description by considering the real risk of classification error. Experimental results demonstrate the effectiveness of the proposed method.The last but not the least, we design a system for tobacco stem detection and tobacco leaf classification. The whole system framework and each function modular are analysis in detail. The related software is also designed by using C++. In order to meet the engineering requirements, our proposed methods are applied to the system with some necessary and rational optimization.