Study On Nitrogen Nutritional Diagnosis Of Cold Region Rice Leaves Based On Hyperspectrum

The yield and quality of rice were related to food security and national economy of our country,and nitrogen was one of the essential nutrients for the growth and development of rice.According to the principle of high investment and high returns,farmers applied a large amount of nitrogen fertilizer,which resulted waste and loss of nitrogen.With the rapid development of agricultural modernization,hyperspectral technology has been widely studied in the field of crop nutrition diagnosis.In this study,Longdao23 and Longdao20 were chosen as the test cultivars,and the nitrogen content of rice leaf was predicted using hyperspectral technology,which provides technical support and theoretical basis for the rapid detection of nitrogen content in rice leaf and the precise fertilization management during rice growth.The experiment was carried out in Rice Research Institute of Fangzheng county,Heilongjiang province in 2015.6 nitrogen fertilization rates were applied for the two cultivars in our experiment,which was carried out at tillering,jointing,heading and maturition growth stage.Hyperspectral images of rice leaves were collected by the hyperspectral platform of Headwall company in the United States,and nitrogen content of living rice leaves were measured by the plant nutrition meter of Zhejiang TOP Instrument company.ENVI software was used to extract the spectral reflectance of rice leaves under 400-1000 nm.The spectral characteristic and the change trend of reflectance of rice leaves were obtained based on the analysis of the spectral reflectance curves and the red edge position curves of rice leaves under different nitrogen levels.Successive projections algorithm(SPA)and segmented principal components analysis(SPCA)were used to select the hyperspectral characteristic bands of rice leaves in all the periods.After reduced the dimension by SPCA,the characteristic spectral parameters were constructed by SPCA combining the correlation analysis(CA).Several regression analysis estimate models which include simple regression analysis,multiple stepwise regression analysis(MSRA),and multiple regression analysis(MRA)have been built based on the full band hyperspectral data,SPA characteristic bands and SPCA characteristic spectral parameters for testing and screening.From the objective performance of the model,the total spectral-MSRA model was the best of Longdao23 in the tillering stage,jointing stage,heading stage,maturation stage,and also the best of Longdao20 in the jointing stage,and the SPCA-CA-MRA model was the best of Longdao20 in the tillering stage,heading stage,maturation stage.From the practical point of view,the total spectral-MSRA model was high accuracy,but with a large number of bands,which was difficult to be used in practice.SPA-MSRA model was slightly low accuracy,but with few bands and its accuracy met the requirements,which was high feasibility.Using SPCA-CA-MRA model,although the computational process was complex,but the model was rich,high precision,and the same feasible.