水稻叶片色素含量近红外光谱估测模型研究  

Estimation of Leaf Pigment Concentration in Rice by Near Infrared Reflectance Spectroscopy

作　　者：姚霞[1] 田永超[1] 倪军[1] 张玉森[1] 曹卫星[1] 朱艳[1] 

机构地区：[1]南京农业大学/国家信息农业工程技术中心/江苏省信息农业高技术研究重点实验室,南京210095

出　　处：《分析化学》2012年第4期589-595,共7页Chinese Journal of Analytical Chemistry

基　　金：教育部新世纪优秀人才支持计划(No.NCET-08-0797);国家自然科学基金(Nos.30900868;30871448);国家863计划(No.2011AA100703);江苏省创新学者攀登计划(No.BK2008037);江苏省科技支撑计划项目(No.BE2010395);江苏省自然科学基金(Nos.BK2008330;BK2010453)资助

摘　　要：以不同品种类型和不同施氮水平的水稻(Oryza sativa)叶片近红外光谱信息为基础,运用逐步多元回归法(Stepwise multiple linear regression,SMLR)、主成分回归法(Principal component regression,PCR)、偏最小二乘法(Partial least square,PLS)和BP神经网络法(Back-propagation neural network,BPNN),建立了水稻叶片中叶绿素a(Chl a)、叶绿素b(Chl b)、叶绿素a+b(Chl a+b)和类胡萝卜素(Car)的近红外预测模型。结果显示,利用8000～4000cm!1波段范围的一阶导数(First derivative,FD)建模效果最佳。其中,基于PLS的预测模型效果最好;4类近红外色素模型的内部交叉验证误差分别为0.251,0.063,0.305和0.073;外部交叉验证的误差RMSEP分别为0.335,0.123,0.302和0.072,表明的预测效果较好。因此,可以基于近红外模型对水稻叶片色素含量进行快速测定。By using the nent regression （PCR） the near infrared reflect techniques of stepwise multiple linear regression （SMLR）, , partial least square （PLS） and back-propagation neural ne ance spectrometry（NIRS）-based models were established for principal compotwork （BPNN）, the estimation of chlorophyll a（Chl a）, Chl b, Chl a＋b and carotene（Car. ） concentration in rice cultivars growing under varied nitrogen rates, which would help with dressing fertilization in rice production. The results showed that the optimum spectral-pretreatment method for the estimation of Chl a, Chl b, Chl a＋b and Car. was the first order derivate spectra with best wave number of 8000-4000 cm-1. The best method was PLS with coefficient of determination for calibration （RC2） all over 0. 8 except Chl a＋b model, while the best principal component factors were all 8. The ：root mean square error of cross validation for Chl a, Chl b, Chl aff-b and Car. model are 0. 251, 0. 063, 0. 305 and 0. 073 with root mean square error for prediction of external validation as 0. 335, 0. 123, 0. 302 and 0. 072, respectively. The results showed that NIRS model of Chl a, Chl b, Chl a＋b and Car. for the analysis of rice leaves were excellent. The NIRS model can he used to estimate leaf pigment concentration for rice.

关 键 词：水稻 叶片色素 近红外光谱 最佳因子数 偏最小二乘法 

分 类 号：S511[农业科学—作物学]