Color Image Segmentation Of Cucumber Leaf In Complex Background And Disease Diagnosis

With the rapid development of modern science and technology, information technology has become an important part of the development of agriculture. Facilities for agricultural which use information technology, especially artificial intelligence technology make traditional agriculture gradually get rid of the shackles of natural thus transform agricultural production and improve agricultural production conditions and the efficiency of agricultural production.In this paper, computer vision technology as a means of combining digital image processing and neural network technology achieve the color segmentation of the cucumber leaf images based on natural background and automatic extraction technology of cucumber’s disease. Furthermore feature extraction of three diseases made some initial progress.Firstly, a natural background color image segmentation algorithm on health cucumber blade is proposed. Anisotropic diffusion method evolved from the partial differential ideologically. Using it’s nature of larger diffused in a small gradient area and smaller diffusion in a larger gradient area protect the edges of the image information while removing image noise. According to the nature of anisotropic diffusion, after dithering and median filtering in the original color image processing then on which using anisotropic diffusion can zone the same scale on the image and protect the edge information in a certain degree at the same time. After that watershed segmentation can solve the problems of over-segmentation to a certain extent. Meanwhile watershed segmentation will produce a lot of small area but automatic seed region can solve it by the combination of the similar color areas.Finally, in accordance with the scale of the last merger get the final segmentation of cucumber leaves. The results show that target area can be segmented better while adding anisotropic diffusion in the pretreatment.Secondly, using improved BP neural network based on genetic algorithm extract and calculating the feature of cucumber disease leaf lesion in the complex background. According to shortcomings of the BP algorithm which is the low efficiency and can not guarantee the optimal parameters of the network structure and weights, this paper combined with genetic algorithms which have the global random search capability and can find the global optimal solution quickly and efficiently in the complex, multi-peak vector space proposed the method suitable for real-coded genetic operators to optimize the BP neural network. The experimental results show that this method can extract the cucumber disease lesion very well in the natural background. Finally, ten texture features and three geometric characteristics is selected from the extracted disease area and find features which can be able to identify different debases by calculating the anglicizing them.Because the fast processing speed and taking up less memory of this paper algorithm, it is very suitable for diagnosis cucumber diseases on embedded platform. It also has certain practical significance to promote the computer vision and artificial intelligence technology in the application of agricultural facilities.