作　　者： ; ; ;

机构地区： 暨南大学理工学院光电信息与传感技术广东普通高校重点实验室

出　　处： 《计算机与应用化学》 2011年第5期518-522,共5页

摘　　要： 采用近红外(NIR)漫反射光谱,建立土壤有机质的NIR光谱分析的偏最小二乘(PLS)模型,比较和选择多元散射校正(MSC)和Savitzky-Golay(SG)平滑用于改善NIR光谱分析的预测能力。把SG平滑模式由原来的117种扩充为483种,并构建了SG平滑模式与PLS因子数联合优选的化学计量学平台。对于单独(或联合)做MSC、SG平滑预处理以及未做光谱预处理的5种情形,分别建立土壤有机质NIR光谱分析的PLS模型。最优的情形是先后做SG平滑和MSC预处理,其中最优的SG平滑参数为5次多项式、3阶导数、21点平滑,对应的PLS因子数、预测均方根偏差(RMSEP)、预测相关系数(R_p)分别为5,0.246(%),0.883,大幅度优于未做光谱预处理的模型预测效果。从而表明,SG平滑和MSC的组合优选可以显著改善土壤有机质的NIR分析效果。所构建的化学计量学平台和方法框架是改善NIR分析能力的有效途径。 Using near infrared（NIR） diffuse reflectance spectroscopy,partial least squares（PLS） models of the NIR spectroscopy analysis for soil organic were established.Multiple scatter correction（MSC） and Savitzky-Golay（SG） smoothing were compared and selected for improving the prediction ability of NIR spectroscopy analysis.The SG smoothing mode was expanded from the original 117 kinds to 483 kinds,and the chemometrics platform for the optimization of SG smoothing mode combined with PLS factor was built up.For five cases of individual use（or joint use） of MSC,SG smoothing preprocessing or no preprocessing,PLS models of NIR spectroscopy analysis for soil organic were established respectively.The optimal case was one preprocessed successively by SG smoothing and MSC.In this case,the optimal smoothing parameters were 5^（th） degree polynomial,3^（rd） order derivation,21 smoothing points,and the corresponding PLS factor,RMSEP,R_p were 5,0.246（%）,0.883,respectively.It was significantly better than the prediction effect of the model without any preprocessing.The results indicated that combination optimization of SG smoothing and MSC can obviously improve the model prediction effect of NIR analysis for soil organic matter.The chemometrics platform and the methodological framework here were the effective way to improve the ability of NIR analysis.

关 键 词： 土壤有机质 近红外光谱 偏最小二乘法 平滑 多元散射校正

领　　域： [机械工程] [理学] [理学]