Study On Nutritional Diagnosis Of Nitrogen Stress In The Cold Rice Based On Hyperspectral Imagery

Heilongjiang province is the best rice producing areas in China, the cultivation of fine varieties and to explore the measures of nutrient management science is to achieve high yield japonica condition. Nitrogen nutrition of rice is one of the main agronomic traits of their growth status of rice leaf color, can directly reflect the nitrogen content, and indirectly reflects the rice water, fertilizer and other internal physiological needs. Through the detection of rice nitrogen nutrition status of precision irrigation, precision fertilization and nutrient management measures, has great promoting effect on Rice Yield in Heilongjiang province. So by measuring the nitrogen nutrition of rice is of great significance to analyze the growth in rice in Heilongjiang province and to predict rice yield in cold region in the work.In this paper, combined with the "863" project "micro UAV remote sensing information acquisition and nutrient management technology" and "postdoctoral science research topic for micro UAV rice nutrient multispectral remote sensing diagnosis technology research based on" etc.. This experiment in Heilongjiang city of Harbin province are Rice Research Institute of the 4 rice varieties, with the sun as the experimental object. According to the experimental field of rice planting distribution, set up 6 fertilizer gradient(never nitrogen, serious lack of nitrogen, nitrogen deficiency, the amount of nitrogen, nitrogen, severe nitrogen), each level has 4 times, a total of 24 district.This experiment using hyperspectral USA produced HeadWall push broom camera application platform, the outdoor experiments, acquisition of visible light and near infrared hyperspectral imaging system(400~1000nm) of rice canopy hyperspectral image in the range of AA3, at the same time using the continuous flow analyzer for determination of leaf nitrogen content. Due to the high spectral data, and the collected image noise, so how to effectively extract the spectral data is very important. This paper uses the median filtering algorithm denoising of the original hyperspectral image, and then the accurate extraction of rice canopy spectral reflectance; analysis of red edge spectral reflectance curves obtained by the technique of rice rice spectral position variables, determine the sensitive wave band range of 520~ 750 nm in visible band; combined with the rice leaf nitrogen concentration indoor analysis, take analysis on the correlation between leaf nitrogen content and leaf spectral parameters, to determine the sensitivity of wavelength, and builds a detection model based on the spectral parameters of nitrogen content of rice canopy nutrient level; using a classification model based on BP neural network. The results show that the correlation analysis, through the analysis of the data and rice canopy spectral reflectance data of chemical nitrogen content of leaves, the characteristic wavelength of rice canopy was 581 nm, 671 nm, 683 nm, 707nm; According to the different levels of nitrogen on rice canopy at jointing stage, heading stage is established based on BP neural network, prediction accuracy level of nitrogen at jointing stage, heading stage reached 96% to 90%; the single band(581nm), a dual band(533Nm, 581nm) and the ratio of difference vegetation index, two three band ratio vegetation index to establish prediction model of nitrogen the estimation accuracy of R2 as a single band of 0.7328, 0.7735 and 0.7592 dual band, three bands 0.8695 and 0.8682, RMSE prediction model for dual band 0.9554 and band 0.6806, three 0.8209 and 0.8195. A comprehensive analysis of three band ratio vegetation index prediction the prediction effect is better.The experiment of different application of rice canopy spectral characteristics of cold nitrogen levels were studied, the results of this study can provide high spectral technology in crop nutrition fast, non-destructive diagnosis provides a certain reference value.