Research On Sunflower Leaf Disease Diagnosis Based On Imaging Identification

Sunflower is an important economical and oil plant, with great development potentialin both domestic and overseas market. However, plant diseases are facing rather serioussituations as soon as sunflowers are concerned, which results in sharp output reduction ofsunflowers. Therefore, to correctly identify the disease type of sunflower and carry outpreventions is an important issue for us to deal with. Traditionally, when identifying thetypes of sunflower diseases, we basically judge from its appearance with our eyes, wherethere is great subjectivity, limitation and ambiguity. For this reason, to develop a computervision and image identification system which is able to simulate human vision and gobeyond its performance as well as to apply the system to the diagnose sunflower diseasehave become an urgent demand of modern agriculture.We has selected three common diseases, respectively, bacterial leaf spot, black spotand downy mildew, as our research targets in the thesis and developed and designed asystem of diagnosing leaf diseases for sunflowers based on image identification. First, wecarry out de-noising processing on the images of sunflower diseases collected under thenatural light by vector median filtering method. Subsequently, we carry out diagnoses as towhether there is diseases based on the image of sunflower’s leaves. In the thesis, we haveadopted characteristic parameter derived from G component as a support for theidentification method of the model of vector input and diagnose whether there is diseasesbased on the images to be tested. As for the abnormal leaves, we carry out diagnosis anddecide whether there is disease by means of gray co-occurrence matrix and support vectormachine. Second, if the image to be tested pictures disease leaves, by adopting thesegmentation of self-decided threshold, we carry out precise segmentation over the diseasespot. When the segmentation is finished, by adopting the opening calculation and closedcalculation in the morphology, we remove independent spots, burrs and small holes fromthe images of disease spot after fine segmentation, eliminating the image noise; Third,Directed at the features of the disease plant after segmentation, we have extracted colorfeature and grain features of the disease spot by utilizing color matrix and gray levelco-occurrence matrix in the thesis. By means of intensive research over the parameters offeatures, we have selected9parameters as proves to identify different diseases. Finally, by adopting one to one polling strategy, we decide that, the vector multiple segmentationmodels are able to make more effective identification diagnosis on the leave disease ofsunflowers.We have taken MATLAB as out platform for the system and developed a diagnosingsystem for the leave disease of sunflowers based on image identification by utilizing GUItool kit. After times of texts, we have found that, the system is able to identify three typesof sunflower diseases effectively, respectively, bacterial leaf spot, black spot and downymildew, which basically meets the requirement raised for the design in the thesis.