CLASSIFICATION OF RICE BROWN SPOT DISEASE SEVERITY BASED ON PROBABILISTIC NEURAL NETWORK

摘要

The spectral reflectance of rice leaves at different disease severity of rice brown spot was analyzed in this paper. On the basis of the analysis, we used three methods, which are resampling, the principal component analysis based on resampling datasets and the first derivative spectra, to deal with the spectra datasets. Three quarters of the data were the training subset to train the neural network, and the remainings were the testing subset. The result demonstrated that the reflectance at the blue and red regions increased as the rice brown spot disease became more serious, however, the reflectance at the region of green, NIR and SWIR decreased as the disease severity increased . It also indicated that the classification precision of the third method was best,which was principle component analysis based on the first derivative spectra. Its classification precision was as high as 80% when the SPREAD was 5. This research made it possible for farmers to manage and control the disease of plant on-time and effectively.