Research on optimal predicting model for the grading detection of rice blast

Abstract

Rice blast is a worldwide disease of rice that is an important reason for the reduction of rice yield. In this paper, “Lingliangyou 268” was selected as the research object. The spectral data were measured by a Landmark Spectrum instrument. The spectral characteristics of the original spectrum, derivative spectrum and logarithmic spectrum of different grades of rice blast were studied. A new method for rice blast grading based on sensitive bands was proposed. Then, the method of system clustering method, BP neural network and probabilistic neural network were used to establish the rice blast classification prediction model, respectively. Comparing the three models, the classification effect based on probabilistic neural network is the best. In the training samples, the logarithmic spectral classification accuracy is 97.8%. In the test samples, the logarithmic spectral classification accuracy is 75.5%.