Research Of Corn Nitrogen Nutrition Mechanism Based On Hyper-spectral Data

Nitrogen nutrition status remains in the spotlight in nutrition-diagnosis of crop. A fast, accurate and dynamic method to diagnose nitrogen nutrition status in large acreage is essential for corn growth vigor evaluation, production prediction and agriculture management.In this research, two study areas respectively located in Changping and Changchun are investigated and multiple spectral parameters are derived from the hyper-spectral data of the corn canopy of the two study areas. Further, based on stepwise regression analysis, quantitative relation between the corn nitrogen nutrition index (NNI) and hyper-spectral parameters is established, and the hyper-spectral parameters which have distinct correlation with the corn NNI as well as the predictive equations are presented. The results show that three hyper-spectral parameters, i.e. NIR/NIR, REIP and NDRAVI, are highly correlated with the corn NNI. Path coefficients of NIR/NIR and REIP to the corn NNI respectively are calculated to be -0.14942 and -0.35218, which indicate an indirect effect of the two hyper-spectral parameters on the corn NNI. The path coefficient of NDRAVI to the corn NNI is 1.41549, revealing that NDRAVI is highly sensitive to the corn NNI. The correlation index of the multi-variant regression model for the distinct hyper-spectral parameters and the corn NNI is 0.91, with fitting error of less than 0.1, RMSE of 0.06016, F value of 167.727 and P value of 0.0045.Short of nitrogen or less will result in some change of crop physical ecology, but this shift in a lot of time is very small. There is more obvious for remote sensing data in a specific period. Many factors affect crop nitrogen. By comparison of different effects on estimating corn leave nitrogen status using kinds of hyper-spectral parameters, the study selected some hyper-spectral parameters which were potentially sensitive to small changes of corn leave nitrogen, and used dynamic fuzzy neural network (DFNN) which would be the selected hyper-spectral parameters and some influencing factors (CWSI,Pb and pH ) as input, corn nitrogen nutrition index(NNI) as output, and the experiment data which were obtained from three natural corn farmlands of the Changchun region as study objects in 2009 to discuss the intelligent method of diagnosis of corn nitrogen deficit. The results showed that comparing with traditional BP neural network forecasting method the presented forecasting method had better adaptive ability and could give better forecasting result(training time of 500 and RMSE of 0.00156. Combining dynamic fuzzy neural network with hyper-spectral remote sensing technology has the potential of improving the estimation accuracy of nitrogen nutrition index.