Ning Yang1, Yong Qian1, Hany S El-Mesery2, Rongbiao Zhang1, Aiying Wang3, Jian Tang3. 1. School of Electrical and Information Engineering, Jiangsu University, Zhenjiang, China. 2. School of Energy and Power Engineering, Jiangsu University, Zhenjiang, China. 3. State Key Laboratory of Rice Biology, China National Rice Research Institute, Hangzhou, China.
Abstract
BACKGROUND: Rice smut and rice blast are listed as two of the three major diseases of rice. Owing to the small size and similar structure of rice blast and rice smut spores, traditional microscopic methods are troublesome to detect them. Therefore, this paper uses microscopy image identification based on the synergistic judgment of texture and shape features and the decision tree-confusion matrix method. RESULTS: The distance transformation-Gaussian filtering-watershed algorithm method was proposed to separate the adherent rice blast spores, and the accuracy was increased by about 10%. Four shape features (area, perimeter, ellipticity, complexity) and three texture features (entropy, homogeneity, contrast) were selected for decision-tree model classification. The confusion-matrix algorithm was used to calculate the classification accuracy, in which global accuracy is 82% and the Kappa coefficient is 0.81. At the same time, the detection accuracy is as high as 94%. CONCLUSIONS: The synergistic judgment of texture and shape features and the decision tree-confusion matrix method can be used to detect rice disease quickly and precisely. The proposed method can be combined with a spore trap, which is vital to devise strategies early and to control rice disease effectively.
BACKGROUND:Rice smut and rice blast are listed as two of the three major diseases of rice. Owing to the small size and similar structure of rice blast and rice smut spores, traditional microscopic methods are troublesome to detect them. Therefore, this paper uses microscopy image identification based on the synergistic judgment of texture and shape features and the decision tree-confusion matrix method. RESULTS: The distance transformation-Gaussian filtering-watershed algorithm method was proposed to separate the adherent rice blast spores, and the accuracy was increased by about 10%. Four shape features (area, perimeter, ellipticity, complexity) and three texture features (entropy, homogeneity, contrast) were selected for decision-tree model classification. The confusion-matrix algorithm was used to calculate the classification accuracy, in which global accuracy is 82% and the Kappa coefficient is 0.81. At the same time, the detection accuracy is as high as 94%. CONCLUSIONS: The synergistic judgment of texture and shape features and the decision tree-confusion matrix method can be used to detect rice disease quickly and precisely. The proposed method can be combined with a spore trap, which is vital to devise strategies early and to control rice disease effectively.