摘要
定量测定小麦叶片叶绿素含量在小麦估产、农情监测等方面具有重要意义。本研究验证高光谱成像技术结合偏最小二乘—最小二乘支持向量机(PLS-LS-SVM)建模方法预测大田冬小麦叶绿素含量的可行性。首先利用所搭建高光谱成像系统以线扫描方式获取大田冬小麦叶片反射光谱,进而得到其立方体图像数据,并在小麦叶片光谱图像上选择感兴趣区域计算出光谱平均反射率值。为保证PLS-LS-SVM模型的鲁棒性和预测稳定性,首先通过PLS方法解决多重共线性问题并将输入变量维数减至4维,然后利用LS-SVM进行训练建模。所建叶绿素含量预测模型的决定系数达R2=0.8459,预测均方根误差RMSEV=0.4370。研究结果表明,基于高光谱成像系统,采用PLS-LS-SVM建立模型用来预测大田冬小麦叶绿素含量是完全可行的。
The main objective was to assess the possibility of predicting biochemical chlorophyll content of field winter wheat using hyperspectral images technology coupling with partial least-squares least square support vector machines （ PLS-LS-SVM） modeling method. Firstly,after the 316 scan line images were acquired,the cube image data was constructed and the region of interest （ ROI） was selected,then after the average reflected pixel intensity has been acquired,the PLS method was used to settle the co-linearity of spectra and to reduce the dimension of variables; then the least square support vector machines （ LS -SVM） was used as the modeling method,the determination coefficient（ R2） between the prediction value and the value obtained using the PLS-LS-SVM modeling method was R^2 = 0. 8459,and the root mean square errors of external validation （ RMSEV） was 0. 4370. The results demonstrated that using our hyperspectral imaging system coupling with PLS-LS-SVM modeling method,we can get a fairly good result. All of these indicated that using the hyperspectral imaging method combined with the relative modeling means,we can predict chlorophyll content of winter wheat precisely.
出处
《农机化研究》
北大核心
2010年第9期170-175,共6页
Journal of Agricultural Mechanization Research
基金
国家"863"高技术计划项目(2006AA10A308
2006AA10A305-1)
国家"十一五"科技支撑计划项目(2007BAD89B04)
关键词
冬小麦
叶绿素含量
高光谱成像
偏最小二乘
最小二乘支持向量机
winter wheat
chlorophyll content
yyperspectral imaging
partial least-squares（ PLS）
least square sup-port vector machines （ LS -SVM）
作者简介
王伟（1975-），男，山东宁阳人，讲师，博士，（E—mail）playerwxw@cau．edu．cn。
