中文摘要：

利用可见/近红外光谱分析技术定量分析叶片生化参数含量时,由光散射效应和叶片厚度等引起的样本间光程长差异会影响校正模型的预测精度。文章提出了一种改进的扩展多元散射校正（EMSC）光程长方法,利用差谱矩阵的主成分代替物质纯谱,对实测光谱精确建模,根据实际需要,减去相应干扰因子,得到校正光谱。选择叶绿素含量基本相同,厚度不同的16个叶片样本以及叶绿素含量和厚度均不同的32个样本分别使用改进的EMSC方法校正,对处理前后光谱变异系数和主成分贡献率进行比较和分析,并比较了预处理前后的模型预测精度。结果表明,采用改进的EMSC预处理方法能够有效地校正光散射效应和叶片厚度差异导致的光程长差异引起的光谱误差,增强光谱数据对浓度的灵敏度,可提高预测模型的精度。

英文摘要：

Vis/NIR spectroscopy technology is capable of analyzing the content of biochemical parameter in folium rapidly and nondestructively. In the process of spectrum analysis, the variations in path-length between different samples exist, with the random light scattering and leaf thickness perturbations, which influence the precision of quantitative analysis model. In order to resolve this problem, an improved path-length correction method based on Extended Multiplicative Scattering Correction is presented. In this paper, firstly the theory of EMSC algorithm is deduced. EMSC method incorporates both chemical terms and wavelength functions to help realize the efficient separation of path-length and interest concentration. Secondly two experiments were implemented to demonstrate the validity of the method. In Experiment 1, sixteen samples of different thickness but almost the same chlorophyll content were selected, and how the path-length affects the spectrum was compared, after EMSC preprocessing, the variable coefficient of spectrum could approach the repeatability error of spectrometer. In Experiment 2, thirty-two samples of different thickness and chlorophyll content were selected. PLS model established using cross validation was employed to evaluate the efficiency of the presented algorithm. Before the preprocessing, the root mean squared error of prediction is 3.9 SPAD with 5 principal components. After preprocessing, the predicted root mean squared error is 2.2 SPAD with 12 principal components. The results indicate that the improved EMSC preprocessing method could exactly eliminate the spectrum difference caused by the path-length variations between different foliums, enhance the sensitivity of concentration and spectral data, and increase the precision of calibrated model.