Abstract: Establishing high-precision SPAD estimation model by machine learning method combined with near-surface remote sensing image parameters based on maize canopy spectral information can provide timely and reliable data of chlorophyll content of leaves in maize canopy. Such a model would facilitate accurate management of nitrogen fertilization in maize fields, reduction of nitrogen fertilizer application, and rapid diagnosis of nitrogen nutrition. The canopy vegetation index from maize jointing to silking stage under six N application levels was obtained by Dajiang four-rotor aerial UAV equipped with digital camera in two years. The correlation between vegetation index and SPAD were analyzed. Univariate regression (UR), multiple regression (MR) and BP neural network algorithm based on principal component analysis (PCA-BP) were established to estimate the SPAD values, and the optimal model was selected and verified after that. The results showed that there were 10 canopy vegetation indices that were significantly correlated with SPAD at the silking stage of maize, and that the normalized redness intensity (NRI), blue-red ratio index (BRRI), difference vegetation index (DVI) and normalized pigment chlorophyll ratio index (NPCI) had higher correlations of above 0.80. The linear, logarithmic, exponential, and power function models of the four indices were constructed. Among those models, the power function model established by NPCI was the best one with the determination coefficient R² of 0.748. The accuracy of the model based on the PCA-BP neural network was the highest (R²=0.818), followed by multiple stepwise regression model, and the univariate regression model had the lowest accuracy. The verification results showed that the estimated value of SPAD based on PCA-BP neural network model was the closest to the measured value with R² of 0.830, RMSE of 0.542, nRMSE of 0.89%, and the prediction effect was the best. In conclusion, the maize canopy SPAD estimation model based on PCA-BP neural network has high accuracy and can provide a new method for the estimation of maize canopy SPAD based on UAV image parameters.

Key words: maize, SPAD, canopy vegetation index, soil nitrogen,  , PCA-BP neural network