Research On Image Recognition Automatically Of Cucumber Disease

Vegetables as an important cash crops in the crop in China, account for a large proportion of China’s agricultural production. Hebei Province is the country’s vegetable production and transferred out of the province. Vegetables of Hebei accounted for one-third in eight agricultural products wholesale market in Beijing and two-thirds in the Beijing-Tianjin-summer market.Cucumber account for56.50%of the total cultivated area, accounting for56.54percent of the total output, but the cucumber annual yield losses due to diseases as high as20%-30%. Currently, most farmers rely on visual observation of the diagnosis of diseases, often resulting in misdiagnosis.Therefore, quickly and accurately identifying the cucumber disease and taking timely and effective measures become to the key technologies of cucumber disease prevention and control. In this paper, researching on image of cucumber disease by digital image processing technology, achieve disease automatic identification,and make up for the lack of experts and professors who can’t go to the on-site guidance.In this paper, the extensive investigation on the Ministry of Agriculture Vegetable point of contact Dingzhou vegetable base cucumber producing, achieve the goal that it can identify common diseases of cucumber, by computer image processing technology. Integrated use of digital imaging, biometrics, pattern recognition and artificial intelligence technology, research on the methods of cucumber disease image recognition. The main tasks are as follows:(1) Acquisition technology research.In order to improve the accuracy of automatic recognition of cucumber disease and avoid the affect of the light that is too strong or too weak, we study the the cucumber disease image acquisition technology.According to the experts,collecting3000leaf images of10common cucumber diseases in fine weather between4:00-6:00am, by the portable image acquisition device(2) Disease image preprocessing methods.By comparing various methods of image preprocessing, histogram equalization and fast median filtering method are elected to use the image for image enhancement. Improving dynamic range and contrast of the image, highlighting the details of the image, and laiding the foundation for image segmentation. Then Otsu threshold segmentation method,and the edge operator method are used for image segmentation of the disease. Finally,processing the segmented image in the voids and burrs by morphological operations, in order to make the target image more specific.(3)Feature extraction and selection of research.According to the different characteristics on the color, texture, shape of different cucumber disease leaf,a total of15features were extracted from the disease image after processing. Accounting to qualitative analysis on the respective characteristic parameters, the paper ultimately chose eight eigenvalues that the mean H, R mean, G mean, S energy, entropy, contrast, the mean of complex shape and circular mean, and combine the features into a feature vector to represent all the knowledge of the original image.(4)Disease Recognition.Parameters to construct the BP neural network classifier based on the selected features.The2000images are selected as the training samples, the1000images are used to identify,and the average recognition rate is93.8%.Compared with experienced farmers recognition results, the results show that recognition of the vast majority of diseases with higher accuracy than the recognition accuracy of the farmers.Experimental results show that, the improved fast median filtering algorithm greatly improve the speed of automatic identification of cucumber disease; select H mean, R mean disease image characteristic parameters as a cucumber disease image recognition feature vector is feasible. Provides a theoretical foundation and scientific basis for the further development of commercially valuable plant diseases automatic diagnosis system.