高光谱成像技术的玉米叶片氮含量检测模型  

Detection Model of Nitrogen Content in Maize Leaves Based on Hyperspectral Imaging

作　　者：王丽凤[1] 张长利[1] 赵越[1] 宋玉柱[1] 王润涛[1] 苏中滨[1] 王树文[1] 

机构地区：[1]东北农业大学电气与信息学院,哈尔滨150030

出　　处：《农机化研究》2017年第11期140-147,共8页Journal of Agricultural Mechanization Research

基　　金：国家"863计划"项目(AA2013102303);黑龙江省博士后科研启动基金项目(LBH-Q13022);东北农业大学科技创新基金项目(yjscx-14003);黑龙江省自然科学基金面上项目(C2015006);哈尔滨市科技创新人才项目(2015RQQXJ020)

摘　　要：应用高光谱成像技术,实现了玉米拔节期叶片氮含量的检测。提取出240个叶片样本的平均光谱反射率数据(400~1 000nm),对原始数据分别进行3种预处理(1stDer、2ndDer、SNV),建立了4种预测模型,包括基于幅值参数(Dλr、Dλy、Dλb)的多种回归模型、全光谱PLS模型、基于连续投影算法(SPA)的PLS模型及基于主成分分析法(PCA)的PLS模型。建模结果显示:基于PCA的PLS模型预测精度最低;全光谱的PLS模型Rc2和RP2分别为0.967、0.821;基于SPA算法的PLS模型R_c^2、R_P^2分别为0.944、0.749,与全光谱的PLS模型预测精度相当,而自变量个数减少了95.07%。基于幅值参数的多元回归模型其预测结果虽与基于全光谱的PLS模型有些许差距,但模型简单,运算量最小,适用于对精度要求不高的场合。Visible and near infrared （Vis-NIR） hyperspectral imaging system was carried out to rapidly detect the nitrogen（N） content in maize leaves. The data of average spectral reflectance, which were acquired by analyzing hyperspectral information（400- 1000nm wavelength） of 240 leaf samples within region of interest （ROI）, were preprocessed by three different methods, including First derivative （ 1 ^st Der）, Second derivative （ 2^nd Der） and standard normal variate （SNV） , to improve the signal to noise ratio. Several estimation models have been built based on different parameters using different algorithms for comparison, including ： （ 1 ） Using three amplitude parameters （ Dλr . Dλy , Dλb ） to establish a single variable regression model and multivariate linear regression model （2） a nitrogen estimation model based on partial least squares （PLS） in the whole wavelength region of 400 - 1000nm; （3） Among 203 wavelengths, only 10 wavelengths were selected by successive projections algorithm （SPA） as the effective wavelengths for N prediction. Based on these effective wavelengths, partial least squares （PLS） calibration model was established for the determination of N content. ; （4） Using principal component analysis （PCA） to extract the 11 main components which were applied to establish 1^st Der-PCA-PLS model. A nitrogen estimation model based on PCA with PLS has the lowest prediction accuracy and prediction results are not feasible. The results indicate that the optimal prediction performance is achieved by PLS model in the whole wavelength region of 400 - 1000nm, which has a correlation coefficient of 0. 967 and a root mean squares error of predicted of 0. 024. The SPA-PLS model can provide a very close prediction result while the calibration computation has been significantly reduced and the calibration speed has been accelerated sharply. This study provibes valuable information for rapid and non-destructive nitrogen detection in crops. For simple ex

关 键 词：高光谱参数 玉米叶片 氮 偏最小二乘法 连续投影算法 主成分分析 

分 类 号：S123[农业科学—农业基础科学]