Research On Remote Recognition System Of Apple Leaf Diseases Based On Andorid Platform

Apple is the major fruit cash crop, and the apple diseases control plays an extremely important role in promoting local economic development. For image analysis technique needs specialized image acquisition devices, and analyzes image in the laboratory, it is difficult to meet the requirements of timely and convenient recognition. This paper took the apple leaf spot, rust, mosaic as the research objects, studied remote recognition of apple leaf diseases based on Android platform, the main contents and conclusions are as follows:(1) Studied the image preprocessing of apple leaf diseases and the method of lesion site segmentation. Using gray linear transformation with three pieces to expand image contrast, then took Bilateral filter to remove leaf tomentum on the H, S, V channel. Converted the image to RGB color space, sharpen image with Laplace filter. At last, morphological Top-hat transformation was used to reduce the impact of light. We will get the preprocessed image. Compared with channel integration of R×R-G×B, R-G-B, R-B, the experiment proved that 2R-G-B has better effect on removing background. Then, we optimized fuzzy c-means clustering based on genetic algorithm, found this method had better segmentation effect than OTSU method, iteration method and Mean shift algorithm. Meanwhile, how to process the segmented image was discussed, including removing vein, filling the hole of the background.(2) Researched the extraction method of color feature, texture feature and shape feature of lesion site. Finally, we selected first moment of H component, first moment of S component, the ratio of R component to G component, the ratio of B component to G component, gray-scale variance, gradient entropy, area ratio, dispersion index of shape complexity as the effective features of lesion site. Then the paper constructed BP network, RBF network, DBN network, and support vector machine for lesion site recognition. Tested the four models with 71 samples, including 24 spot, 17 rust, 30 mosaic images. The experimental results showed that support vector machine reached higher recognition accuracy of 98.60%.(3) We developed remote recognition system of apple leaf disease based on Android platform. Android mobile phone was used to obtain image as the client, and uploaded the image to the server through 3G network. Then the server analyzed image, returned recognition results to client. Considering the requirements of time and accuracy, chose SVM to recognize lesion site, the mean time was 15.7s. The system provides convenient and fast service of recognition and prevention guidance of the apple leaf diseases for farmers.