Study On Rice Chlorophyll Contents Estimation Based On Large Margin Neatest Neighbor Classification

Rice was one of the most important alimentarn crop in China.The production of rice took an important place in agriculture.In the recognition of nitrogen nutrition status, nitrogen was the ingredient of chlorophyll, that was, it was an easy way to realize the recognization of nitrogen nutrition status via estimating the contents of chlorolphyll.Then,the reasonalbe fertilization scheme was laid down. The problems of the decline of rice production and pollution were solved.Therefor, the estimation of the contents of rice chlorophyll had an significant and realistic meaning.In this paper, the results were as follows:Firstly,rice leaf images and chlorophyll contents were obtained.The test object is kongyu131rice.In the field enviroment and white background,the vertical leaf images were obtained.Using CM1000chlorlphyll meter to get the chlorlphyll contents.Secondly,some edge detection operators were intrduced.On the basis of analysing the advantages and disadvantages of the operators,an edge detection algorithm based on multi-strategy fusion technique was proposed.The new algorithm combined median filtering,canny operator, mathematical morphological with minimum bounding rectangle.Compered with the traditional operators,the new algorithm was accurite and robust.Thirdly, in this paper,21kinds of color features were extracted considering that color features can reflect chlorophyll contents.Because of the redundancy of color features,PCA and ICA were used to reduce the dimensions of color features.The results were2principal components and4independent components.Lastly,translating the problem into discrimination learning problem.On the basis of introducing KNN and TKNN,a two-classifications model and a multi-classifications model based on LMNN were build,which could estimate the chlorophyll contents.The experimental results show that the method based on PAC and LMNN can get the best classification accuricy both in two-classifications and multi-classifications.The problem of estimating the contents of rice chorophyll on the basis of image processing was an important part of accurate agriculture.In order to solve the problem,in this paper,image processing and dimension reducing were combined with discrimination learning.This methord was not only a complementary part of the estimation of chlorophyll contents,but a framwork of solving the resemble problems.