Design And Implementation Of Apple-leaf Disease Image Rec-Ognition System

Apple industry is the characteristic industry of Shaanxi province, and also hasimportant significance for the economy. But in recent years, with the prosperous de-velopment of apple industry and the increasing of plant area, the apple production hasbeen influenced gradually by various diseases, and fruit farmer’s income also de-creased. It’s important to eliminate the disease problem happened in large area. So theperfect image processing and intelligent recognition technology is one way to solve theproblem. This paper focuses on three kinds of apple-leaf diseases including spot, rustand mosaic, which are common and highly incidence, analyzes the diseases’ character-istics, extracts the features, designs and develops the recognition system. All of thework provide technical reference to the remote identification of apple diseases. Maincontents and conclusions are as follows:(1) According to the characteristics of images taken under the natural environment,the image preprocessing methods were studied. Improved median filtering method wasbe used to remove the noises and kept the edge information better; histogram equaliza-tion method was be used to enhance the images(The method could make the enhancedimages and have a large dynamic range and high contrast); fuzzy c-means clusteringalgorithm was be used to segment the disease images and got good results.(2) The methods about how to extract features from images were be studied, es-pecially studied the features including colors histogram, colors moments, gray levelco-occurrence matrix, common shapes and Hu invariant moment. The extracted fea-tures were be compared and15of them were be selected as the classification featureparameters. (3) The support vector machine (SVM) learning model was be studied. Selecteddiseases recognition model based on the support vector machine (SVM). The correctmodel parameters were be tested and found.(4) C#and Matlab languages were be used to program and implement all the func-tions of the apple-leaf disease image recognition system. The result of the experimentshowed the average accuracy could be91%. The system could recognize the apple-leafdiseases effectively.