The Research On Auto-Grouping Model Of Flue-Cured Tobacco Leaves Based On Digital Image Processing

As a kind of important industrial crops, flue-cured tobacco is only graded with a manual sensory method in China, which can't overcome the effect of subjective factors. Therefore, the research of an auto-grading system based on digital image processing is highly significant to improve the grading efficiency and accuracy. The primary goal of the research is to offer key data for tobacco grading and establish a recognition pattern model for grouping tobacco leaves. Through digital image processing techniques, the color parameters of the reflection images and transmission images are collected. Then, the grouping model based on the collected information via statistical pattern recognition is established.The main contents and achievements of this thesis are shown as follows:1. Various methods and instruments are applied to conduct the preliminary processing: for reflection image, we use mathematical morphology functions in Matlab to split the background through the way of calling Matlab engine in VC++; Based on that most of the blue vector value of pixels in transmission image were equal to zero, an critical point threshold calculating by area-oriented statistics to segment the transmission image is proposed. On the basis of this research, the respective characteristics and application scopes of some commonly adopted colorimetric models are studied, and designed a new algorithm with the integration of RGB and HSI color system for the description of the color feature of tobacco leaves.2. The colorimetric theory and digital image processing techniques are combined to imitate man's chromatic vision to do the quantitative measurements and descriptions. The RGB and HSI color feature values of reflection and transmission images are collected. To reduce the errors caused by illumination, standard color reference plane is applied to modify the RGB value of reflection images.3. With the assistance of statistical methods, different grading models are established based on RGB, R' G' B', HSI, H' S' I' parameters of reflection and transmission image. Tested with the self-validation and cross-validation, the one with the lowest rate of error is the most superior one. One hundred and thirty sample leaves which are not trained are used to test the model. The successful grading rate reaches 92%.The results indicate the high feasibility of the establishment of a flue-cured tobacco leaves auto-grading model with digital image processing techniques and model recognition methods. In grouping models, a comprehensive application of transmission images as a supplement to reflection images can increase the reliability of grouping, and will be applied in a wide range.