Segmentation And Classification Identification Of Early Tobacco Leaves Disease Spot Based On Computer Vision

In 2018,the overall tax and profit of tobacco industry in China reached 111.51 billion RMB,accounting for about one-tenth of China's total fiscal revenue.But every year,considerable economic losses occur because tobacco diseases cannot be timely prevented and controlled.The common tobacco diseases in central China are brown spot and frog-eye leaf spot of tobacco;and the early symptoms of them are the most difficult ones to distinguish.The inexperienced plant protection personnel and farmers cannot judge those two early diseases correctly,and even employ wrong pesticide.As a result,diseases fail to be stopped in time and lead up to a significant reduction of output and its income.At present,there is no relevant report on the study of early tobacco diseases in China.The incidence rate of Brown spot disease and Frog-eye leaf spot of tobacco disease is fast,and the disease is about one week from early stage to late stage.At present,there are many problems in the research results,such as the inability to segment effectively under the complex background,the detection of diseases only in the middle and late stage,and the reliance on manual inspection.Therefore,an efficient and accurate automatic detection and identification method for early tobacco diseases is urgently needed to solve the above problems.According to tobacco red star of the characteristics of big disease and early frogeye similarity in this paper,a kind of tobacco leaves early disease segmentation and recognition based on computer vision method,effective segmentation and recognition of the two diseases,can be found about 3 days in advance and confirmed that diseases,targeted intervention,the largest extent,reduce the loss.(1)The data of brown spot and frog-eye leaf spot of tobacco were collected in tobacco field.The right time point,the right photo distance and the right photo objects were chosen,so the best effect of the disease image are enabled to achieve.Mean shift smoothing and simple linear iterative clustering are applied to the acquired images.(2)This article through studies EXG super green character segmentation method and threshold segmentation method of HSV color space,found EXG super green in the green noise characteristic of image segmentation method can not effectively break up,threshold segmentation method of HSV color space to poor disease spot area and background contrast can't accurate segmentation under the condition of disease spot area.According to the red star disease and tobacco frogeye disease spot in different historical periods,performance characteristics,will be introduced to the visual significant disease spot segmentation process,based on seeds point selected the significance of detection method,verified the feasibility of disease spot image segmentation under complex background,and compared two kinds of image segmentation method,by experiments on the prototype validate the effectiveness of the method.(3)According to the different manifestation forms of disease spots in each period of tobacco red star disease and frog eye disease,the color characteristics,morphological characteristics and texture characteristics of disease spots were extracted,and a total of 28 dimensional characteristic parameters were obtained.The obtained characteristic parameters were optimized by particle swarm optimization(pso),and the optimal parameter group was selected through multiple experiments.The fitness was 96.68,the cross-validation rate was 93.21%,and the recognition rate of the verification set was 96%.A total of 13 dimensional characteristic parameters were selected.Using the grid search method for SVM parameters optimization,and establish the model of SVM classification,the training set 900 groups,300 groups of test set,a total of 1200 sets of data,the tobacco red star disease and early disease of frogeye recognition rate reached 92%,two kinds of disease early medium night threes be period,a total of 6 class recognition rate reached 96%,the experimental results show that to achieve good recognition effect.