The prediction of rice leaf´s nitrogen content based on leaf spectrum on the heading stage

The change of crops´ nitrogen content can cause the surface of crop leaf and the physiological characteristics of the internal organization to change, thus can cause the spectrum reflection characteristic of the crop leaf to change. In this paper, the amount of fertilizer was controlled, and nitrogen-containing samples of the rice cultivation experiment was conducted to study the relevant relations of the reflection spectrum of the rice leaf and the nitrogen content of the rice leaf in the earing period. BP network and LM neural network, Bayesian neural network are used to set up prediction models of rice leaf´s nitrogen content, and a comparative analysis of the network training situation and the predicted results are carried on. The results show that LM neural network converges faster than BP neural network, and convergence precision of Bayesian neural network is higher than BP neural network´s. In terms of prediction accuracy, LM neural network is the best.