Estimating nitrogen status of rice canopy using hyperspectral reflectance combined with BPSO-SVR in cold region

As special soil background, climate condition and growth period in cold region, it is necessary to find adaptable models to assess rice growth and nutrient situation. This study was conducted to estimate the nitrogen content of rice in different growth stages using rice canopy's hyperspectral reflectance toward the development of precision nitrogen status monitoring. The field experiment was undertaken applying four levels of nitrogen (N) treatment for cultivar 'Daohuaxiang'. The hyperspectral reflectance of rice canopy of tilling stage, jointing stage and heading stage was captured using hyperspectral imaging system in the range of 372-1038 nm covering 128 wavebands. The average spectral reflectance was extracted from five region of interest (ROI) of each sample and a total of 192 groups of spectral reflectance data were obtained. Then, eight vegetation indices were calculated by spectral reflectance. A kind of machine learning method, termed “support vector regression based on binary particle swarm optimization algorithm (BPSO-SVR)” was proposed to predict nitrogen content. The results were achieved by selecting the best subset of input variables and optimizing the parameters 'c' and 'g' of SVR through the method of BPSO-SVR. In this work, we also established traditional prediction models such as partial least square regression (PLSR), principal components regression (PCR) and GA-BPANN. The predictive power of these regression models was compared using R2 (coefficient of determination) and RMSE (root mean square error) of calibration set and testing set. The newly proposed 'BPSO-SVR'method yielded the excellent R2 (0.913-0.949) and the smaller RMSE (0.055-0.127) for fitting nitrogen concentration of rice canopy over three growth stages. The results showed that, the method proposed in this paper for predicting N content of rice canopy in different growth stages was potential for nitrogen status monitoring in cold region.