基于高光谱的甜菜不同生育时期氮素营养参数监测研究

发布时间：2018-08-22 10:17

【摘要】：本研究于2014年与2015年分别在内蒙古赤峰试验站和内蒙古农业大学农场开展了甜菜田间试验。以两年的田间小区试验为基础,在不同氮素水平下,开展基于高光谱技术的甜菜不同生育时期(幼苗期、叶丛生长期、块根膨大期、糖分积累期)氮含量及SPAD值的监测研究。本研究的主要工作和研究成果如下:1.采用ASD高光谱仪获取甜菜不同生育时期的冠层光谱反射率,研究甜菜冠层高光谱响应特征,发现不同生育时期的可见光波段(450-680nm)及近红外波段(760-950nm)的光谱响应存在明显差异;分析经平滑和一阶导数预处理后的冠层光谱与氮含量、SPAD值的相关关系,结果表明:一阶导数光谱对氮含量、SPAD值的敏感波段较多,且相关性得到很大提升。2.以优选光谱参数作为输入变量,研究不同算法(一元回归、PCR、PLSR、SVM)建立的高光谱预测模型对甜菜各个生育时期氮素营养参数预测精度的影响,研究结果表明:在幼苗期,氮含量及SPAD值的最佳预测模型分别为PLSR模型和PCR 模型,其 R2 分别为 0.520、0.552,RMSE 分别为 2.70g/kg、1.99,RE 分别为6.6%、3.5%;在叶丛生长期,SVM模型对氮含量的预测效果最好(R2、RMSE及RE分别为0.443、3.17g/kg、10.1%),以SDr/SDy建立的二次函数模型对SPAD值的预测效果最佳(R2、RMSE及RE分别为0.212、1.85、3.3%);在块根膨大期,氮含量及SPAD值的最佳预测模型均为PCR模型,其R2分别为0.447、0.536,RMSE分别为2.93g/kg、1.81,RE分别为9.7%、3.3%;在糖分积累期,PLSR模型对氮含量的预测性能最佳(R2、RMSE及RE分别为0.344、2.95g/kg、9.0%),以SDb为自变量建立的对数函数模型对SPAD值预测效果最好(R2、RMSE及RE分别为0.324、3.17、5.9%)。本研究通过建立甜菜不同生育时期更为精准的氮含量及SPAD值的高光谱预测模型,从而为甜菜田间的施肥管理提供理论依据,为甜菜氮素营养参数的快速无损监测提供技术支持。
[Abstract]:In this study, sugarbeet field trials were carried out at Chifeng Experimental Station in Inner Mongolia and Farm in Inner Mongolia Agricultural University in 2014 and 2015 respectively. Based on two years field plot experiment, the monitoring of nitrogen content and SPAD value of sugarbeet at different growth stages (seedling stage, leaf growth stage, root swelling stage, sugar accumulation stage) based on hyperspectral technique was carried out under different nitrogen levels. The main work and results of this study are as follows: 1. The canopy spectral reflectance of sugarbeet at different growth stages was obtained by using ASD hyperspectral analyzer. The characteristics of hyperspectral response of sugarbeet canopy were studied. It was found that the spectral responses of visible light wave (450-680nm) and near infrared band (760-950nm) in different growth stages were significantly different. The correlation between the canopy spectrum and the nitrogen content (SPAR) after smooth and first derivative pretreatment is analyzed. The results show that the first derivative spectrum has more sensitive bands to the nitrogen content and the correlation has been greatly improved. The effects of hyperspectral prediction models established by different algorithms (single regression PCRX PLSR-SVM) on the prediction accuracy of nitrogen nutrition parameters in each growing period of sugarbeet were studied with the selected spectral parameters as input variables. The results showed that: 1. The best prediction models for nitrogen content and SPAD value are PLSR model and PCR model, respectively. The R2 of RMSE was 0.520 / 0.552g / kg respectively 2.70 g / kg ~ (1.9) RE was 6.6g / kg / kg ~ 1.9RE = 6.6and 3.55.The prediction effect of SDr/SDy model on nitrogen content was the best (R _ 2 / R _ (MSE) and RE = 0.443n ~ 3.17 g / kg _ (10.1%), respectively). The quadratic function model established by SDr/SDy had the best effect on predicting SPAD value (R2N RMSE and RE, respectively). It was 0.2121.85% (3. 3%). The best prediction model for nitrogen content and SPAD value is PCR model. The R2 of RMSE was 2.93 g / kg 路kg ~ (-1) 路L ~ (-1), respectively, and that of SDb model was the best (R _ (2) N _ (RMSE) and RE = 0.3442.95 g 路kg ~ (9.0%). The logarithmic function model with SDb as independent variable had the best effect on SPAD prediction (R2RMSE and RE were 0.3243.175.9%, respectively). In this study, a more accurate hyperspectral prediction model of nitrogen content and SPAD value in different growth stages of sugarbeet was established, which provided theoretical basis for fertilizer application management in sugarbeet field, and provided technical support for rapid non-destructive monitoring of nitrogen nutrition parameters in sugarbeet.
【学位授予单位】：内蒙古农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S566.3

【相似文献】