Remote sensing estimation of nitrogen content in summer maize leaves based on multispectral images of UAV

Abstract: At present, the remote sensing monitoring system composed of UAV platform and multi-spectral camera has achieved some results in agriculture. However, there are few attempts to use UAV multi-spectral image to estimate crop nitrogen. Based on this, the multi-spectral images were acquired by UAV and leaf nitrogen contents (LNC) were measured in the National Precision Agriculture Base in 2017 to carry out the estimation research on the nitrogen content of summer maize leaves in this paper. Crop LNC estimation model were constructed mostly based on single spectral variable in traditional research methods, but the model constructed by single spectral variable were easy to be saturated, while excessive selection of variables may lead to over-fitting. The stepwise regression method is a multivariate regression analysis method, which is simple and easy to perform. The obtained regression equation has fewer variables and retains the advantages of the most significant important variables. Therefore, the stepwise regression model is used to estimate the nitrogen content of leaves. In this study, 48 sets of sample data were obtained in 3 growth stages. Firstly, the data were preprocessed, and the preprocessing of UAV multispectral data included image mosaic, radiation calibration and geometric correction. Secondly, 15 spectral variables were selected to analyze the correlation with LNC, and then the spectral variables sensitive to LNC at different growth stages were screened out. Finally, the backward stepwise regression method was used to analyze the change of estimation accuracy under different variables, and the spectral variables to estimate LNC in different growth stages were determined to achieve higher precision monitoring of LNC in summer maize. It can be found that: 1) In the 3 growth stages, the green band reflectance and the green normalized difference vegetation index(GNDVI) constructed by the green band had strong correlation with LNC, indicating that the green band can perform the inversion of LNC in summer maize; 2) The optimal soil adjustment vegetation index(OSAVI), soil adjustment vegetation index(SAVI) and LNC were highly correlated in trumpet stage and filling stage, which proved that the selection of spectral variables reflecting soil factors in early and late growth stage of summer maize can improve inversion accuracy of the nitrogen content. Considering the evaluation index and simple practicability of the estimation model, 5 spectral variables were selected according to the adjusted determination coefficient (R2adj) in the trumpet stage, 6 spectral variables were selected in the anthesis silking stage, and 5 spectral variables were selected in the filling stage to construct the model. In the trumpet stage, the R2, root mean square error (RMSE)and normalized RMSE(nRMSE) of the estimation model were 0.63, 27.63% and 11.62%; In the anthesis silking stage, that were 0.64, 20.50% and 7.80%; In the filling stage, that were 0.56, 31.12% and 12.71%; It can be found that the application of UAV multi-spectral remote sensing image data can well monitor the spatial distribution of LNC in field-scale summer maize, and provide spatial decision service information support for corn field precision management.