基于成像光谱技术的寒地玉米苗期冠层氮含量预测模型

Forecasting model for nitrogen content of maize canopy during seedling stage in cold region based on imaging spectral technique

为了探索寒地玉米冠层氮素含量,以不同氮素水平下玉米大田试验为基础,利用高光谱成像技术探讨苗期玉米冠层光谱,通过相关矩阵法选择植被指数的变量,并依据叶片氮素含量与植被指数的相关性,建立玉米冠层氮素含量预测模型。结果表明:根据玉米冠层高光谱图像,选择与各波段相关性较强的525、566、700、715、895 nm作为植被指数的变量,构建与氮素含量相关性强的植被指数归一化植被指数NDVI(normalized difference vegetation index)、归一化光谱植被指数NDSI(normalized difference spectral index)、比值光谱指数RSI(ratio spectral index)、差值光谱指数DSI(difference spectral index)。以与叶片氮素含量相关性较高的植被指数为自变量,建立单变量、多变量回归预测模型。采用单变量NDVI二次函数回归模型作为0、50 kg/hm^2施氮量下玉米冠层氮素含量预测模型,其R^2分别为0.719、0.803。在100 kg/hm^2施氮量下玉米冠层氮素含量的预测模型为3变量回归模型,其R^2达到0.657。用置信椭圆F检验法检验预测模型,其F值均小于F0.05,估测值与实测值间R2分别是0.724、0.798、0.655,标准误差RMSE分别为0.156、0.140、0.156 mg/g,表明实测值和估测值间的差异不明显,预测模型可用。

Heilongjiang is a big province of corn production, where corn is widely planted. Because of the influence of soil and light, the research about corn canopy at stage of seedling with imaging spectral technique is less. In recent years, the direct analysis of spectral data and the monitoring modeling of vegetation index have become effective methods in the research of imaging spectrum technology on crop nutrient and growth analysis. So this paper was focused on the canopy at seedling stage of maize. To explore the nitrogen content of maize canopy in cold region, 2 methods were used to analyze the spectral data of canopy image. The experiment was carried out in Fangzheng county, Harbin city. The tested corn variety was Heyu20. The fertilization gradient of each test region was 0, 50 and 100 kg/hm2 nitrogen. This experiment used the imaging spectrometer to collect image and the German AA3 analyzer to measure the corn ammonium nitrogen content. To ensure the integrity of the image, this paper chose to extract the image directly. In this experiment, we chose the high correlation band with each band as the variable of vegetation index. The bands were the 43rd band （525 nm）, 57th band （566 nm）, 102nd band （700 nm）, 10th band （715 nm） and 168th band （895 nm）. And then bands were brought into the vegetation index about RSI （ratio spectral index）, DSI （difference spectral index）, NDSI （normalized difference spectral index） and NDVI （normalized difference vegetation index）. Under 50 kg/hm2 nitrogen application rate, the content of nitrogen in maize leaves had a higher correlation with NDVI, RSI （715nm, 700 nm）, DSI （700nm, 566 nm）, and NDSI （715nm, 700 nm）. And under 0 and 100 kg/hm2 nitrogen application rate, the nitrogen content of maize leaves had a higher correlation with NDVI, RSI （895nm, 700 nm）, RSI （715nm, 700 nm）, DSI （700nm, 566 nm）, DSI （895nm, 700nm, 525 nm） and NDSI （715nm, 700 nm）. This paper established the single variable and multivariable forecasting model of nitrogen element content in maize canopy by using vegetation index and nitrogen content. The function included power function, exponential function, logarithmic function, linear function and quadratic functions. This paper tested the accuracy with the confidence ellipse F test model. Under the nitrogen application rate of 0 and 50 kg/hm2, the effect of the prediction model with NDVI was the best, and the R2 values were 0.719 and 0.803 respectively. When the nitrogen applieation rate was 100 kg/hm2, the multivariate prediction effect was better, and the R2 reached 0.657. Using the F test to examine the forecast model, its F-measure was less than Foos. The R2 values between predicted values and measured values for the nitrogen content of maize canopy under different nitrogen application levels were 0.724, 0.798 and 0.655 respectively and the RMSE （root mean squared error） values were 0.156, 0.140 and 0.156 mg/g, respectively; the forecast model was available and the prediction model could be used. In this paper, the prediction model of nitrogen content in maize canopy has good applicability, and can provide support for the application of micro-UAVs （unmanned aerial vehicles） in agriculture.