滤光片型近红外仪器模型传递的研究

模型传递问题是近红外光谱分析技术中解决数据通用性的关键问题。文章以玉米籽粒近红外光谱图(检测其中水分含量)为例，考察了模型传递的问题。使用斜率截距算法，直接校正法和目标因子分析等算法，在5台滤光片型近红外仪器上实现了模型传递，并比较了各种方法的模型传递效果。研究表明，直接校正法的模型传递效果最好，4台从仪器的平均传递差异度为7.01%。文章还研究了标准样品数量对模型传递效果的影响。作为转换集的标准样品数目越多，模型传递效果越好，一般有20个标准样品就能达到稳定的效果。当转换集小于20时，直接校正法的传递效果急剧下降，而标准样品数量对斜率截距法和目标因子分析法的影响不明显。

Research on Calibration Transfer of NIR Filter Spectrophotometer

Calibration transfer is an important issue to building up universal and comparable performance of spectrometer data in near infrared spectral analysis technology. Methods of slope/bias correction, direct standardization (DS), and target factor analysis (TFA) were used for the calibration transfer among five NIR filter spectrophotometers using maize as the samples. The effects of three calibration transfer methods were compared. The DS method has the best performance. The average calibration transfer difference of DS is 7.01%. This study also relates to the dependence of calibration transfer on the number of standardization samples. It was proven by experiment that the results of calibration transfer will be better as the number of samples is increased and will be generally stable when there are twenty standardization samples. However, the effect of calibration transfer attained by DS is degraded sharply when the number of standardization samples decreases to be below twenty. Moreover, slope/bias and TFA are not sensitive to the number of standardization samples.