An improved BP neural network model for estimating Cd stress in rice using remote sensing data

Monitoring heavy metal stress in rice is significant for agricultural production management and food security. Remote sensing offers an undamaged and efficient approach to detect the crop and soil contamination. In this study, an improved BP neural network for predicting the accumulation of the total cadmium (Cd) in rice was proposed by using the genetic algorithm (GA) and the simulated annealing (SA). To establish the model, the spectral parameters of the sensitive factors, including the normalized difference moisture index (NDMI), ratio vegetation index (RVI), and enhanced vegetation index (EVI), were taken as the input data set. The results indicated that the SA-GA-BP model performed well for predicting the Cd with R²=0.98, RMSE=0.081, ME=0.981. The comparison between SA-GA-BP and BP neural network also showed that the accuracy of the SA-GA-BP model was better than the BP model. It was demonstrated that the use of SA and GA to improve the BP neural network model was feasible and more suitable to predict the Cd.