Research On Automatic Identification System Of Tobacco Diseases

Tobacco is one of the major cash crops of China, and is an important source of tax revenue. According to statistics, China had planted 1837 acres of flue-cured tobacco in 2014, industrial and commercial of the tobacco industry taxes reached 1.05176 trillion yuan, the annual of total 911.03 billion yuan turned over to finance. Quality is the production of high quality tobacco cigarette raw material, and is one of the important premises to ensure the quality of cigarette. However, the tobacco in the growth process due to the influence of climate, soil, technology and other factors, easily susceptible to tobacco diseases hazards, reduce tobacco production, affecting the quality of tobacco. The traditional tobacco disease diagnose depends on manual identification and laboratory testing. It has low efficiency, poor real-time, can not be large-scale applications and the other issues. Therefore, how to quickly and accurately identify the tobacco diseases is a serious problem.In this paper, a variety of tobacco diseases was collected under field environment. Interactive segmentation algorithm was used to accurately extract the tobacco image spots. With color moments, Hu moments and GLCM got tobacco diseases color, contour and texture feature values. Genetic neural network was used to establish the diagnosis of tobacco diseases recognition model. By the moving client and the server, the tobacco diseases identification system was applied to tobacco production. Research for the tobacco diseases recognition provided an accurate, convenient and efficient solution.This paper proposed a combination of Otsu and GrabCut interactive segmentation algorithm for extracting. First, mobile phone was using to capture images of tobacco diseases in the field environment. Then, the tobacco diseases picture was converted from RGB color space to HSV color space, extracted V component diagram. Again, V component diagram of the tobacco diseases was used Otsu segmentation method and calculated for each lesion location and size information. Finally, the position information initialized GrabCut information lesion area by analyzing the characteristics of the region had obvious lesion, the lesion accurately segmented region to obtain tobacco diseases lesion segmentation map.This paper proposed a multi-feature fusion recognition algorithm of tobacco diseases. By color moments, contours and GLCM extracted color, contour and texture features of tobacco diseases. Using genetic algorithm to optimize BP neural network BP neural network to obtain the optimal initial weights and thresholds, shorten training time. In order to improve the recognition accuracy of tobacco diseases the multi-feature was fused by genetic neural network to establish automatic recognition tobacco diseases diagnosis model.This paper designed a tobacco diseases recognition mobile client. Tobacco diseases identification system built on a mobile client and server. When found tobacco diseases in the field users took pictures and uploaded images to server. Server automatically recognized the type of tobacco diseases and fed back to the user’s mobile clients including recognition result, disease characteristics and control methods. Users based on the feedback results to guide the production of tobacco.In this paper, to establish a mobile client as input, customer service as output, through a multi-feature fusion built the genetic BP neural network model of tobacco diseases. Tests showed that:this method could effectively identify eight kinds of tobacco diseases such as Red star, Climate spots, Viral, Wildfire and Angular spot, and the average recognition accuracy was 92.5%.