论文部分内容阅读
以MPA型傅里叶近红外光谱仪和WQF-400N型傅里叶近红外光谱仪上检测的奶粉光谱为原始光谱,利用积分法选择不同的积分区间对原始光谱进行积分,模拟不同分辨率下的光谱。采用相同的预处理方法——Savitzky-Golay平滑和一阶微分,利用偏最小二乘(PLS)回归对奶粉中的脂肪含量建立模型。分别比较两台仪器所建模型的相关系数(R),预测均方差(fRMSEP),平均相对误差(E)三个预测参数,得到两台仪器分别在分辨率为16 cm-1和64 cm-1时建模效果最好。研究结果表明,利用积分光谱,能起到压缩数据,提高模型预测精度的作用;同时说明对于具体样品的特定指标要选择合适的分辨率,在减少工作量的同时获得最佳的定量分析结果,而单纯的追求高分辨率是没有意义的。
The spectra of milk powder detected by MPA FT-IR spectrometer and WQF-400N FT-NIR spectrometer were used as the original spectra. The integrals were selected by integral method to integrate the original spectra and simulate spectra under different resolutions . The same pretreatment method, Savitzky-Golay smoothing and first-order differentiation, was used to model the fat content in milk powder using partial least squares (PLS) regression. The correlation coefficients (R), the prediction mean square error (fRMSEP) and the average relative error (E) of the two models were compared respectively to obtain the prediction accuracy of the two instruments at resolutions of 16 cm-1 and 64 cm- 1 modeling best. The results show that using integral spectrum can compress the data and improve the prediction accuracy of the model. At the same time, it is necessary to select the appropriate resolution for the specific index of a specific sample, to obtain the best quantitative analysis results while reducing the workload, The pursuit of high-resolution simply does not make sense.