The Image Segmentation For Wheat Leaf Disease Base On Markov Random Field And K-means Clustering

With the gradual deterioration of the global climate, the environment has become increasingly vulnerable, and the extreme weather phenomena is getting more and more frequently, thus the outbreak of crop diseases shows a rising trend. Our country is a big agricultural country, so our lives and economic development are closely relevant to agricultural security, how to prevent crop diseases efficiently and accurately has became a major issue today. It is the widespread application of computer that makes various pattern recognition technology be mature gradually, hence, using image processing technology to process crop diseases and to extract feature parameters could achieve the goal of identifying disease artificially, which provides a new method for the diagnosis and prevention of disease.The steps of disease recognition include image preprocessing, disease image segmentation, feature extraction and recognition classification of the lesion. The efficiency and accuracy of identification of disease greatly depends on the validity of image segmentation. Leaf disease is the main object of study of image segmentation, which appears obviously when the crop are infected by pathogen. The purpose of the image segmentation for wheat leaf disease is to divide the lesion region under complex background, but due to the complexity of the background (non-controllable illumination, background feature diverse, overlapping leaves, weeds and so on), image segmentation from complex background environment has been the technical bottleneck of disease recognition.The method of image segmentation proposed in this thesis is based on the a Markov random field and K-means clustering, which is aimed at the image segmentation of typical symptoms disease of wheat leaf (leaf rust, leaf blight) under complex background. Using this method to test a multiple sets of image, the test have been shown to be of great robustness and high segmentation accuracy. Finally, it has designed segmentation and calculation system of leaf disease in order to extract leaf lesion under complex background and calculate the relative lesion area on leaves. This thesis contains the following contents:(1)This thesis analyzed the existing methods of image segmentation and summarized the applicability and limitations of different segmentation algorithms. Then the theoretical foundation of Markov random field was described detailedly, especially analyzed the implementation of iterative condition mode (ICM) based on random field theory in the image segmentation process.(2)The segmentation of lesion image under complex background. The test images processing with medium value filter then imported into computer with a size of 800*800 JPG format. The image lesion segmentation was realized gradually:Firstly, in order to remove the influencing factors from complex background such as dirt and shadow, using Markov random field and mathematical morphology to process original color image, then the leaf area with lesion were segmented successfully; Secondly, using space clustering method to divide lesion area from leaf image; Finally, respectively reckoning pixel area according to the leaf area and lesion area, then the relative lesion area was calculated as a measurement index of severity of disease.(3)The verification experiment of the method of segmentation. The experiment took a sets of wheat leaf images with rust disease and blight disease as test objects, and the pixels under the condition of Photoshop manual segmentation were deemed to standard value. The results show that the algorithm misclassification rate is 3.53% on average. Compared with the traditional method OTSU, the algorithm proposed in this thesis could segment the lesion area more completely and it has a higher accuracy and a better applicability.(4)Leaf disease segmentation and computation system was developed by Matlab platform, which is aimed at the intelligent segmentation and calculation of crop disease. The system provides a convenient and practical tool for the latter disease recognition and diagnosis. The calculation of the relative lesion area provides data to support the development of pathology, such as quantitative assessment of the extent of disease,measurement of pathogen infection, genetic research and disease control.