Research On Identify Technologies Of Apple's Disease Based On Image Analysis

This paper used the images of apple pests which captured by mobile phone to identify the pests and diseases of apple, and the emphasis on apple leaf involved five kinds of diseases such as spot, rust, round spot, mosaic, yellow leaf disease, etc. We specially focused on the diseases whose place is larger and more serious damaged. And all of the research based on image analysis of Apple's early disease identification and classification. Main contents and conclusions are as follows:(1)Focusing on the complex type of noise in low-resolution images which is shot under natural conditions using a mobile phone, and in-depth study of the image pre-processing of the denoising problem, we set up a complete pre-treatment process: The image sections of the linear method with three gray adjust, expanded the dynamic range of gray, used Top-Hat transform to eliminate the effect of light; Treated the color image with the anti-color operation, and enhanced image through the RGB channel separation, median filtering, channels fusion; Finally, we used improved between-class variance threshold method and morphological operations and region labelling methods such as the pathological parts of diseased leaves collected on the enhanced image RGB component map, the results show that the red (R) component is the ideal segmentation, it can be used to segment a lesion region.(2) Studying the feature extraction on apple disease image, we achieved good results after making experiments through the three aspects features of the test images such as color, texture and shape. According to lesion shape and lesion Hu invariant moments of shape feature extraction, H-variance combined H-S lesion characteristics extraction as a lesion of the color histogram feature, the box-counting dimension of texture features used to extract from lesions. Finally, we chose 8 categories from the 27 test characteristics as optimized parameters of the classification.(3) Analyzing four kinds of pattern recognition method, we used the BP neural network model for disease identification, the results which after image recognition of the five kinds of apple diseases showed that: the average accuracy rate is 92.6%, which meet the needs of Apple's remote diagnostics. (4) Using VC++ and Matlab mixed programming to realise the main system-related function modules of Apple's disease identification based on image analysis of cell phone cameras, to provide conveniences for the majority of technicians and farmers with apple production safety, the main disease identification, treatment and reasonable medicine and other technical advisory services platform.Paper conclusions will promote the technique of computer image analysis and the application will have some references in other crop diseases of remote diagnosis.