基于高光谱技术的烟草含氮化合物估测模型研究

发布时间：2017-12-27 09:14

本文关键词：基于高光谱技术的烟草含氮化合物估测模型研究　出处：《山东农业大学》2016年博士论文　论文类型：学位论文

【摘要】：烟草氮素是影响其内在质量最为重要的营养元素,准确及时的烟草氮素营养判断是保证氮肥合理施用、保证烟叶质量的基本前提。研究快速、无损的烟株氮素营养信息提取方法,对于准确判断烟株氮素营养水平,指导烟田氮素管理决策,促进氮肥的合理施用,实现烟株正常生长发育和烟田生态系统的健康发展均具有重要的理论和现实意义。本研究选择烤烟相对集中的产区,在自然环境(自然光源、自然噪声环境等)下,研究高光谱特征参数与烤烟烟叶总氮含量的相关关系,并建立估测模型,预测烤烟叶片总氮、烟碱含量,及时获取烤烟生育过程中总氮、烟碱含量变化信息,为实现大面积遥感预测烤烟氮素营养丰缺评估、指导烤烟追肥奠定基础。主要研究结果如下:1)烟草冠层叶绿素含量估测模型:利用烟草冠层在663nm、1173 nm和1670nm处反射光谱的一阶微分作自变量,得到叶绿素含量估测模型Y=25.640+6487.507×R’663-20299.859×R’1670-13751.203×R’1173,决定系数达0.602。基于位置变量和面积变量所建的叶绿素估测模型预测精度高于一阶微分和植被指数两种参数所建模型,并且以红边幅值和红边面积为自变量所建模型Y=30.732+11195.95×Dr-256.608×SDr,模型决定系数为0.839;经检验,叶绿素实测值与估测值相关系数为0.916,RMSEP为1.6449,模型精度较好,可以预测现蕾期烟叶叶绿素含量。2)烟草叶片叶绿素含量估测模型:以现蕾期高光谱特征值λ682、λ494、λ1932为自变量建立的多元线性估测模型为Y=44.27-2233.55λ682+2721.33λ494-61.56λ1932,F值38.5247,p值为0.0000,决定系数R2=0.5463。利用实测值对估测模型进行检验,拟合方程的预测值与实测值相关系数为0.8354;RMSEP为3.8209;RE为8.92%;模型具有良好的拟合效果,决定系数R2达到0.6980。3)烟草叶片总氮含量估测模型:以NDVI(573,440)建立的单变量总氮含量估测模型Y=9.8752X NDVI(573,440)^-1.1849,决定系数R2=0.6756;拟合方程的RMSEP为3.1926;RE为11.78%。模型精度较好,可以用来预测现蕾期烟叶总氮含量。以高光谱特征值SDr/SDb、NDVI(573,440)、NDVI(660,440)、NDSI(FD700,FD690)为自变量建立的多元估测模型优选Y=30.5397+14.0959xNDSI(FD700,FD690)-32.0771x NDVI(573,440)+5.2260xNDVI(660,440)+0.9540x SDr/SDb,估测模型的R2=0.631;拟合方程预测值与实测值的RMSEP为7.5077;RE为33.87%,模型精度较好。4)烟草叶片烟碱含量估测模型以现蕾期高光谱参量(SDr-Sy)/(SDr+Sy)建立单变量估测模型Y=2.2427/(1+EXP(6.1463-4.7476x(SDr-Sy)/(SDr+Sy)))预测烟叶烟碱含量,其值与实测值的相关系数达到0.8148,均方根误差(RMSEP)为0.2140,相对误差(RE%)15.47;用估测模型Y=-4.2628+5.8974x(SDr-Sy)/(SDr+Sy)-1.3260x(SDr-Sy)/(SDr+Sy)2预测烟叶烟碱含量,其值与实测值的相关系数达到0.7955,均方根误差(RMSEP)为0.2272,相对误差(RE%)14.42。多元估测模型方程Y=-0.4753+1.6982x(SDr-Sy)/(SDr+Sy)-0.1616x Rg/Ro预测的烟叶烟碱含量,与实测值的相关系数达到0.8841;预测方程的均方根误差(RMSEP)为0.2335,相对误差(RE%)15.47。经过模型检验,现蕾期建立的单变量估测模型Y=2.2427/(1+EXP(6.1463-4.7476 x(SDr-Sy)/(SDr+Sy)))、Y=-4.2628+5.8974x(SDr-Sy)/(SDr+Sy)-1.3260x(SDr-Sy)/(SDr+Sy)2和多变量估测模型Y=-0.4753+1.6982x(SDr-Sy)/(SDr+Sy)-0.1616x Rg/Ro都可用于现蕾期烟草叶片烟碱的预测。
[Abstract]:Tobacco nitrogen is the most important nutritional element affecting its internal quality. Accurate and timely nitrogen nutrition judgement is the basic premise to ensure the rational application of nitrogen fertilizer and ensure the quality of tobacco leaves. Study on rapid and nondestructive tobacco nitrogen nutrition information extraction method, the accurate judgement of tobacco nitrogen level, nitrogen in tobacco field guide management decisions, promote the rational application of nitrogen fertilizer, which has important theoretical and practical significance to realize the healthy development of the normal growth of tobacco and tobacco field ecosystem. This research chooses the relative concentration of the flue-cured tobacco producing areas, in the natural environment (natural light and natural noise environment), correlation between the total nitrogen content of flue-cured tobacco and high spectral characteristic parameters, and establish the estimation model of prediction of total nitrogen and nicotine content of flue-cured tobacco leaves, the contents of total nitrogen and nicotine to obtain timely change information from the growth of tobacco in the process, lay the foundation for the realization of large area remote sensing prediction of tobacco nitrogen nutrient abundance assessment, guidance of topdressing for tobacco. The main results are as follows: 1) estimation of Canopy Chlorophyll content of tobacco model: using tobacco canopy as independent variables in first-order differential 663nm, 1173 nm and 1670nm reflectance spectra, obtained the chlorophyll content estimation model of Y=25.640+6487.507 * R '663-20299.859 * R' 1670-13751.203 * R '1173, decision coefficient was 0.602. The position variable and variable area chlorophyll estimation model of the prediction accuracy is higher than the first-order differential vegetation index and two parameters based on the model, and the red edge amplitude and area of red edge as the independent model of Y=30.732+11195.95 * Dr-256.608 * SDr model, the coefficient of determination was 0.839; after the examination, and estimate the value of the correlation coefficient is 0.916 the measured values of chlorophyll, RMSEP is 1.6449, the accuracy of the model is good, can predict the chlorophyll content of budding leaf. 2) the estimation model of chlorophyll content in tobacco leaves: the multivariate linear estimation model based on the hyperspectral eigenvalues of lambda, lambda 494, and lambda 1932 as Y=44.27-2233.55, 682+2721.33, lambda, 494-61.56 and 1932, F 38.5247, P value 0, and determination coefficient R2=0.5463. The estimated model is tested by the measured value. The correlation coefficient between the predicted value and the measured value of the fitting equation is 0.8354, RMSEP is 3.8209, RE is 8.92%, the model has good fitting effect, and the determination coefficient R2 reaches 0.6980. 3) the estimation model of total nitrogen content in tobacco leaves: the Y=9.8752X NDVI (573440) ^-1.1849, the coefficient of determination R2=0.6756, the RMSEP of the fitting equation is 3.1926, and the RE is 11.78%, based on the total variable nitrogen content of NDVI (573440). The precision of the model is good, and it can be used to predict the total nitrogen content of tobacco leaves at the bud stage. SDr/SDb and NDVI with high spectral characteristic value (573440), NDVI (660440), NDSI (FD700, FD690) as the independent variable to establish the estimation model of multivariate optimization Y=30.5397+14.0959xNDSI (FD700, FD690) -32.0771x NDVI (573440) +5.2260xNDVI (660440) +0.9540x SDr/SDb, the R2=0.631 estimation model; fitting equation of the predicted value and the measured value of RMSEP 7.5077; RE is 33.87%, the accuracy of the model is good. 4) the nicotine content of tobacco leaves estimation model at bud hyperspectral parameter (SDr-Sy) / (SDr+Sy) a single variable estimation model of Y=2.2427/ (1+EXP (6.1463-4.7476x (SDr-Sy) / (SDr+Sy))) to predict nicotine content, the correlation coefficient value and the measured value reached 0.8148, the root mean square error (RMSEP) for 0.2140, the relative error (RE%) estimation model with 15.47; Y=-4.2628+5.8974x (SDr-Sy) / (SDr+Sy) -1.3260x (SDr-Sy) / (SDr+Sy) 2 Prediction of nicotine content, the correlation coefficient value and the measured value reached 0.7955, both Fang Genwu (RMSEP) was 0.2272, the relative error (RE%) 14.42. Multivariate prediction model equation Y=-0.4753+1.6982x (SDr-Sy) / (SDr+Sy) -0.1616x Rg/Ro predicted nicotine content in tobacco leaves, and the correlation coefficient between the measured values and the measured values reached 0.8841. The root mean square error (RMSEP) of the prediction equation was 0.2335, and the relative error (RE%) 15.47. After the model test, the single variable Y=2.2427/ model to estimate the squaring period established (1+EXP (6.1463-4.7476 x (SDr-Sy) / (SDr+Sy))), Y=-4.2628+5.8974x (SDr-Sy) / (SDr+Sy) -1.3260x (SDr-Sy) / (SDr+Sy) and 2 multivariate estimation model of Y=-0.4753+1.6982x (SDr-Sy) / (SDr+Sy) -0.1616x Rg/Ro can be used to prediction of the budding period of nicotine in tobacco leaves.
【学位授予单位】：山东农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S572

【相似文献】