Research And Application Of Image Recognition Method Of Cucumber Disease In Greenhouse

With the development of computer technology and image technology,automatic identification of crop diseases has become a research hotspot.The application of image recognition technology in crop disease diagnosis,in addition to solving the labor shortage,it is also of great significance to improve the output and quality of crops,reduce costs and promote the development of modern agriculture in China.Based on the research results of the 2017's project of Sichuan education department,"The research and design of automatic image recognition subsystem for greenhouse cucumber disease,a intelligent diagnosis system for crop disease which is based on image processing",this thesis proposes the key algorithms of image preprocessing,disease spot segmentation,feature extraction and disease recognition of three common diseases of Cucumber in greenhouse with digital image processing,pattern recognition and software technologies.And the identification system of cucumber disease is constructed.This thesis includes the following parts:(1)The RGB image is adjusted to a gray image without color information interference by piecewise linear transformation,and then the gray level in the image is widened or compressed by histogram equalization calculation to improve the clarity of the image.Finally,an adaptive median filter algorithm is proposed based on the traditional median filter,which can filter impulse noise with higher probability.The experimental results show that the method in this paper can complete image preprocessing well,and keep the image clear and detail edge while removing noise.(2)the method of histogram threshold segmentation was proposed to select the appropriate threshold value to separate the diseased spot image from the normal image.Experiments show that this method can effectively remove the complex background,reduce the interference to the image of disease spot,and obtain the complete image of disease spot.(3)A new feature extraction technology is constructed to convert the RGB space into HIS space,and the color histogram is improved to design a cumulative histogram to extract 12 color feature values.In MATLAB,the gray co-occurrence matrix was constructed through functions to extract 8 texture feature values.The principal component analysis method was used to select 9 representative values from the 20 characteristics as experimental parameters.The results show that the classification accuracy of the optimized feature parameters is higher than that of the original feature parameters.(4)Based on the actual situation of this study,the experimental samples were randomly selected several times.Two methods of pattern recognition,BP neural network and support vector machine(SVM),are compared and tested.Through the "one-to-one" method,with the help of LibSVM toolbox under MATLAB,the classification model is designed by radial basis kernel function to complete the disease recognition.The experimental results show that the average recognition rate of three diseases of cucumber leaves by SVM is 96.39%,and the recognition accuracy is high.(5)Based on the Android development platform IDE,the design and implementation of the cucumber disease recognition system was completed with the help of JDK,Eclipse,MySQL,Tomcat,Matlab and other tools.The system provides farmers with the functions of disease diagnosis and disease information inquiry,which has obvious practicability and universality.