Estimation of rice neck blasts severity using spectral reflectance based on BP-neural network

Estimation of rice disease using spectral reflectance is important to non-destructive, rapid, and accurate monitoring of rice health. In this study, the rice reflectance data and disease index (DI) were determined experimentally and analyzed by single wave correlation, regression model and neural network model. The result showed that raw spectral reflectance and first derivative reflectance (FDR) difference of the rice necks under various disease severities is clear and obvious in the different spectral regions. There was also significantly negative or positive correlation between DI and raw spectral reflectance, FDR. The regression model was built with raw and first derivative spectral reflectance, which was correlated highly with the DI. However, due to rather complicated non-linear relations between spectral reflectance data and DI, the results of DI retrieved from the regression model was not so ideal. For this reason, an artificial neural network model (BP model) was constructed and applied in the retrieval of DI. For its superior ability for solving the non-linear problem, the BP model provided better accuracy in retrieval of DI compared with the results from the statistic model. Therefore, it was implied that the rice neck blasts could be predicted by remote sensing technology.