Liu Wei-yang, Peng Jie, Dou Zhong-jiang, Chen Bing, Wang Jia-qiang, Xiang Hong-ying, Dai Xi-jun, Wang Qiong, Niu Jian-long
Guang Pu Xue Yu Guang Pu Fen Xi. 2017 Jan;37(1):156-61.
Plant canopy pigment concentration is a critical variable for agricultural remote sensing due to its close relationship to leaf nitrogen content. The aims of this study were to: (1) compare the prediction performances on chlorophyll, chlorophyll-a and b, and carotenoid concentration in jujube leaf at canopy scale between partial least squares regression (PLSR) and support vector machine (SVM), (2) develop quantitative models to estimate pigment concentration in jujube canopy using hyperspectral data and provide theoretical and technical support for rapidly, non-destructive, less expensive and eco-friendly measuring the concentration. Results from correlation analysis showed that jujube canopy pigment concentration correlated strongly with hyperspectral data. What’s more, the hyperspectral data was better correlated by chlorophyll and chlorophyll-a than chlorophyll-b and carotenoid. Results of independent samples tested in predicting performance indicated that both of the PLSR and SVM models could effectively estimate pigment concentration, however, with different prediction precisions. Additionally, the precision of SVM outperformed PLSR for predicting chlorophyll and carotenoid. Whereas chlorophyll-a and chlorophyll-b were better predicted using PLSR than SVM. Compared among all the pigments’ prediction precisions with corresponding optimal inversion models showed that prediction precisions on chlorophyll, chlorophyll-a and carotenoid were superior to chlorophyll-b. The determination coefficients and residual prediction deviation from predicting chlorophyll, chlorophyll-a and carotenoid were higher than 0.8 and 2.0, respectively, while the mean relative error values were lower than 13%. And the corresponding values from predicting chlorophyll-b were 0.60%, 20.79% and 1.79% respectively.
由于植物冠层色素浓度与叶片氮含量密切相关，因此它是农业遥感中的一个关键变量。本研究的目的是：（1）比较偏最小二乘回归（PLSR）和支持向量机（SVM）在冠层尺度上对枣树叶中叶绿素、叶绿素a和b以及类胡萝卜素浓度的预测性能；（2）利用高光谱数据建立定量模型来估算枣树冠层色素浓度，为快速、无损、低成本且环保地测量色素浓度提供理论和技术支持。相关性分析结果表明，枣树冠层色素浓度与高光谱数据密切相关。此外，高光谱数据与叶绿素和叶绿素a的相关性优于叶绿素b和类胡萝卜素。预测性能的独立样本测试结果表明，PLSR和SVM模型都能有效估算色素浓度，但预测精度不同。此外，SVM在预测叶绿素和类胡萝卜素方面的精度优于PLSR。而PLSR在预测叶绿素a和叶绿素b方面比SVM更好。在所有色素的预测精度与相应最优反演模型之间进行比较表明，叶绿素、叶绿素a和类胡萝卜素的预测精度优于叶绿素b。预测叶绿素、叶绿素a和类胡萝卜素的决定系数和剩余预测偏差分别高于0.8和2.0，而平均相对误差值低于13%。预测叶绿素b的相应值分别为0.60%、20.79%和1.79%。
