The Design Of The Machine Vision Classification Model Based On Image Processing On Corn And Weeds

Because of complex background of field and improper selection of classifier parameters, method of the recognition of crop and weed based on machine vision which have been developed also haav some disadvantages, such as low efficiency, accuracy act. In order to solve the problem and improve the efficiency and accuracy of recognition model, classification model of weed identification from corn based on machine vision has been studied in this paper. The main research contents of this subject include:(1) Contrary to complex background of field, by means of analyze and compare, Put forward the image brightness correcting by homographic filter. Extract 2G-R-B component image in RGB color model, and then segmented by Otsu method, final accurate extract image of corn and weed.(2) The feature of corn and weed picture has been analyzed. We get 5 parameters of shape characteristics:entropy,energy,homogeneity,contrast,correlation.The gray level co-occurrence matrix is briefly discussed,Five texture features were extracted based on R;entropy,energy,homogeneity,contrast,relevance.The fractal dimension of the image is obtained based on "Blanket algorithm",Obtain the fractal dimension’s range of corn is 1.1532～1.2026.the fractal dimension’s range of weed is 1.0337～1.0889.(3) The basic situation of SVM and BP neural network has been discussed. SVM classification model and BP neural network classification model has been constructed based on the characteristic parameters. To realize the automatic optimization of parameters of SVM classification model, Gauss function is used as kernel function of the SVM and the parameters has been optimized by using the niche genetic algorithm which is based on sharing function. The results show that SVM classifier is more efficient and accurate,The average recognition rate reached 95.5%.