Estimating Nitrogen Content of Corn Based on Wavelet Energy Coefficient and BP Neural Network

In order to estimate the nitrogen content of corn in natural environment timely and accurately, we developed a method to determine the nitrogen status of corn based on wavelet energy coefficient and BP neural network. hyperspectral reflectance (350-1300nm) was performed by wavelet transform using Dubieties 5 wavelet function and nine level wavelet coefficients of spectral reflectance. Wavelet energy coefficients and nitrogen content were used as the independent and dependent variable of regression model, respectively. Of all the models, wavelet energy coefficient F2 as the independent variable achieved the best with R2 of 0.905. A BP neural network model based on a five wavelet energy coefficients (F1~F5) with relative R2 was used as input parameters, and output parameter of the output layer was nitrogen content. The result showed that an optimum BP neural-network prediction model has 5-4-1 network architecture with R2 of 0.932 and root mean square error (RMSE) of 0.097. The result indicates that the model with wavelet energy coefficient and BP neural network can extract characteristic variables from hyper spectra. Compared with the regression analysing model, it can improve the accuracy of estimation of corn´s nitrogen content.