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: 《分析化学》 2012年第6期920-924,共5页
摘 要: 利用移动窗口偏最小二乘(MWPLS)和Savitzky-Golay(SG)平滑方法优选土壤总氮的近红外(NIR)光谱分析模型。从全部97个土壤样品中随机选出35个样品作为检验集;基于偏最小二乘交叉检验预测偏差(PLSPB),将余下62个样品划分为具有相似性的建模定标集(37个样品)、建模预测集(25个样品)。最优波段为1692~2138 nm,SG平滑的导数阶数(OD)、多项式次数(DP)、平滑点数(NSP)分别为0,6,69,PLS因子数为11,建模预测均方根偏差(M-RMSEP)、建模预测相关系数(M-RP)分别为0.015%,0.931,检验预测均方根偏差(V-RMSEP)、检验预测相关系数(V-RP)分别为0.018%,0.882。其结果可为设计专用NIR仪器提供有价值的参考。 Moving window partial least square(MWPLS) and Savitzky-Golay(SG) smoothing methods were used to optimize near-infrared(NIR) spectroscopic analysis model of total nitrogen in soil.Thirty-five samples were selected randomly from all 97 soil samples as validation set.The remaining 62 samples were divided into modeling calibration set(37 samples) and modeling prediction set(25 samples) with similarity based on the predictive bias of cross-validation of partial least square(PLSPB) model.The optimal waveband was 1692-2138 nm,order derivative(OD),degree polynomial(DP) and number of smoothing points(NSP) of SG smoothing parameters were 0,6 and 69 respectively,PLS factor was 11,the root mean square error of modeling prediction(M-RMSEP) and the correlation coefficient of modeling prediction(M-RP) were 0.015% and 0.931,respectively,the root mean square error of validating prediction(V-RMSEP) and the correlation coefficient of validating prediction(V-RP) were 0.018% and 0.882,respectively.The results could provide valuable reference for designing specialized NIR instruments.