Diagnosis For Main Disease Of Winter Wheat Leaf Based On Image Recognition

This research which based on agricultural production aimed at the rapiddiagnosis of the main diseases of winter wheat leaf, integrated the knowledge andexperiences on winter wheat pathological, digital image processing and pattern recognition.Using Matlab and Java Web technology, this study realized the main diseases diagnosisimage of winter wheat leaf. Researching contents and results were mainly on followingaspects:1. In the aspect of disease images enhancement, analyses and compare with methods ofhistogram equalization, wiener filter and median filter according to the features of winterwheat disease images, we found that histogram equalization treatment effect was not good,but wiener filter, median filter can be better to keep the image edge and detail information,the enhancement effect was satisfactory, among them, median filter has simple operation andis especially effective for salt and pepper noise, so use this median filter method to enhancethose disease image in realization of system.2. In the aspect of disease segmentation, analyses and compare with3methods ofthreshold, morphological gradient and selected ROI, we found that threshold method is morerelies on the user’s subjective judgments, and because of great majority images’ histogramare discrete and irregular, so it has larger differences in selection of threshold. Segmentationmethod of selected ROI is more intuitive and simple than the2other methods, it can alsodirectly implementation of color image segmentation, and the effect was satisfactory,准确率达96%, so use this method of selected ROI to segment scab images in realization of system.3. In the aspect of feature extraction and image recognition, mainly extracted the color,shape and textural features from sample images. The extracted feature vectors which include3color features,5shape features and5textural features were saved in the training set formatstored in a text file. While image recognition mainly used maximum likelihood classificationand support vector machine classification, these2algorithms both have a high rate of correctrecognition:89.7%and87.7%, and the identification of powdery mildew in the correct rateis the highest, reaching98.9%and99.3%.4. Integrated a diagnosis system for main disease of winter wheat leaf, which based onB/S,3layer architecture model of J2EE, MVC and its Struts technology, this system wascompleted with the help of related processing functions of Matlab image processing toolbox,and the integrated development environment of MyEclipse, MySQL database, Tomcat6.xserver. The system includes image processing, lesion recognition and standard lesion image database of3modules, to achieve the user upload disease image, and processed imageenhancement, segmentation, then extract information of upload image scab, classificationcombined with scab database information, finally got the belonging of upload images, and atthe same time the user could also query the related condition and prevention measures of thedisease. In addition, this system backstage database contains a lot of standard spot images,the user can also use the standard library to view spots, understand the relevant informationof all kinds of disease images.