多元散射校正对近红外光谱分析定标模型的影响

Effects of multiplicative scatter correction on a calibration model of near infrared spectral analysis

采用近红外漫反射光谱分析技术,用傅里叶变换型光谱仪对50个烟叶样品采集吸收光谱,采用常用的多元散射校正(MSC)对光谱预处理,通过主成分分析、相关谱等方法比较分析了预处理对光谱分析的影响,用偏最小二乘(PLS)回归法建立近红外光谱与总糖含量的定标模型,用 Leave One Out的交叉检验(Cross Validation)检验定标模型,结果 PLS 因子数由 MSC校正前的 5 降为校正后的 3,RM SECV值仅由0.884 1%降为0.85%。实验证明:对光谱进行MSC预处理能有效减少模型的最佳因子数,简化数学模型,使模型更稳定,更便于传递,但并不能显著减小最优定标模型的预测标准差,即不能显著提高模型的预测能力。

Near infrared diffuse reflectance spectroscopy was used to analyze 50 tobacco samples for total sugar. Principal component analysis (PCA) and correlation analysis were performed to investigate the effect of multiplicative scatter correction (MSC) on spectra analysis. Different calibration models produced by partial least squares (PLS) regression before and after MSC pretreatment were compared by using cross-validation. After MSC pretreatment, the number of PLS factors, dropped from 5 to 3 and the RMSECV merely reduced from 0.8841% to 0.85%. The result shows that MSC can reduce the number of PLS factors needed, simplify and stabilize a calibration model and facilitate model transfer, but can not improve the prediction ability of the model dramatically.