Research On Technology Of Rice Blast Rcognition Based On Support Vector Machines

The research on rice disease recognition based on computer vision is significant to theaccurate understanding of rice blast，guiding agricultural producers controlling diseases，reducing the impact of disease on rice yield and ensuring national food security. In thisthesis，the acute type， chronic type and white point type rice blast color images wererecognized by using support vector machine(SVM)， based on the excellent classificationperformance of SVM， and it is the basic research to study and develop the automaticrecognition system of rice blast.A segmenting algorithm of rice blast color images was discussed based on twoclassifications SVM， in order to improve the accuracy of image segmentation. The differentkind’s of rice blast spots were segmented by two classifications SVM. The imagesegmentation results were compared and analyzed with different classification kernelparameters. The best model parameters of image segmentation SVM were obtained. Theexperimental results show that the algorithm can improve the segmentation accuracy of riceblast color image and the accuracy based on this SVM model was better than OTSU.First， the rice blast color features were extracted and analyzed in RGB， HSI，YCbCr and NTSC color space，and four color features were selected，in order to extract thefeatures of different rice blast spots. Then，shape feature extraction were studied for differentkind’s rice blast spots. The new shape features were defined which are disease spots numberand the ratio of disease spots area and disease spots number， and four shape features wereselected. The simulation results and data analysis show that the eight features can effectivelydistinguish the three different rice blast diseases and can be used as the vector for diseaserecognition multi-class SVM.Different kind’s of rice blast were classified based on the multi-class SVM， by usingthe best combined feature sets， and the best model parameters were obtained. Theexperimental results show that the rice blast recognition algorithm has obtained highrecognizable accuracy based on multi-class SVM， the average correct recognition rate is93.3%， can recognize the disease image of different rice blast.