Dissertation > Agricultural Sciences > Plant Protection > Pest and Disease Control > Crop pests and diseases and their prevention > Economic crop pests and diseases > Tobacco pests and diseases

Tobacco Diseases Auto-Recognition Research Based on Image Processing Technology

Author WangJing
Tutor ZhangYunWei
School Kunming University of Science and Technology
Course Agricultural Electrification and Automation
Keywords tobacco disease image segmentation feature extraction fuzzy recognition
CLC S435.72
Type Master's thesis
Year 2011
Downloads 14
Quotes 0
Download Dissertation

As one of Chinese important economic crops, tobacco occupies a significant position in the national economy. Meanwhile, the tobacco diseases not only threaten tobacco leaves quality and restrict cigarettes quality, but also make huge losses to our agricultural production. Therefore, tobacco diseases prevention and treatment become a urgent problem to be solved. As the traditional tobacco diseases prevention and treatment can not solve the diseases species recognition problem of actual production timely and exactly, the best disease prevention and disposal time may be missed. Therefore, the research uses image processing and recognition technology to distinguish and diagnose crop pest diseases,and as the two main diseases collected in the field which is bad for the tobacco leaves:tobacco blown spot and tobacco wildfire for the objects of study, the image processing and recognition technology are studied in this paper. The major achievements obtained as follows:1.Collecting pictures of tobacco leaves diseases in the tobacco farming demonstration plots by digital cameras, and then at the suggestion of plant protection experts, gathering 80 pictures of tobacco blown spot and tobacco wildfire which are more serious.2.Studying the method of image enhancement. The method mainly including monochrome transformation processing disposal, monochrome histogram equalization and two types of image smooth filters comparison. The experiment proves that median filter algorithm not only has a good effect on noise cancellation, but also reserves the marginal information, which is in favor of regional division of the disease spot. Therefore, median filter algorithm is chosen in this paper.3.Image disease spot segmentation of disease leaves. By comparing and analyzing the traditional image segmentation algorithm and two kinds of color space segmentation algorithm, we can find that based on the H-component of HSI color space, the OTSU segmentation algorithm can effectively separate disease spot area from the leaves. Based on that conclusion, the opening and closing operation functioned by mathematical morphology is used on the segmented binary image. That removes the isolated little dots and achieves the desired segmentation. 4.Feature extraction of disease spots. By analyzing the shape, texture and color characters of disease spot image which divided from the normal part, the appropriate characteristic parameters as a basis for disease identification can be selected.5.Disease identification. According to the characteristic parameters extracted by the fourth step, the fuzzy pattern recognition method that based on fuzzy theory is selected for this paper to identify the disease categories. And the classifier is also designed specially for the tobacco diseases that studied in this paper. First, we building the standard mode space, and then select appropriate membership functions and calculate the approach degree of clustering core and that functions, finally get the good identification based on the nearby principle.This study on one hand provide theoretical basis for the intelligence, automation, diagnostics and prevention of non-destructive of tobacco and the management network basis for remote diagnostics, on the other hand, the study can be directly applied to tobacco disease accurate diagnosis and fast positioning recognition, provide technical support for safety tobacco production, and have important scientific significance and practical value.

Related Dissertations
More Dissertations