基于稳定竞争自适应重加权采样的光谱分析无标模型传递方法

提出了一种基于稳定竞争自适应重加权采样(stability competitive adaptive reweighted sampling，SCARS)的无标模型传递方法。利用有用信息标准即稳定度指数(定义为回归系数除以其标准偏差的绝对值)和传递后的预测均方根误差(root mean squared error of prediction，RMSEP)，选择重要的、受测样参数影响不敏感的波长变量，能够消除或减少不同仪器或测量条件对样本信息反应差异，提高模型传递效果。此外，在该方法中，光谱变量被压缩、降维，从而使模型传递更稳定。采用该方法对谷物的近红外光谱分析模型在不同仪器之间进行传递研究。结果表明，该方法能消除仪器间的大部分差异，较好地实现模型传递效果。与正交信号校正法(orthogonal signal correction，OSC)、蒙特卡罗结合无用信息变量消除法(Monte Carlo uninformative variable elimination，MCUVE)、竞争自适应重加权采样法(competitive adaptive reweighted sampling，CARS)的比较表明，SCARS不仅在传递精度上能取得比OSC、MCUVE及CARS更好的效果，而且能有效地对光谱数据进行压缩，简化并优化传递过程。

Calibration Transfer without Standards for Spectral Analysis Based on Stability Competitive Adaptive Reweighted Sampling

A novel calibration transfer method based on stability competitive adaptive reweighted sampling (SCARS) was proposed in the present paper. An informative criterion, i.e. the stability index, defined as the absolute value of regression coefficient divided by its standard deviation was used. And the root mean squared error of prediction (RMSEP) after transfer was also used. The wavelength variables which were important and insensitive to influence of measurement parameters were selected. And then the differences in responses of different instruments or measurement conditions for a specific sample were eliminated or reduced to improve the calibration transfer results. Moreover, in the proposed method, the spectral variables were compressed, making calibration transfer more stable. The application of the proposed method to calibration transfer of NIR analysis was evaluated by analyzing the corn with different NIR spectrometers. The results showed that this method can well correct the difference between instruments and improve the analytical accuracy. The transfer results obtained by the proposed method, orthogonal signal correction (OSC), Monte Carlo uninformative variable elimination (MCUVE) and competitive adaptive reweighted sampling (CARS), respectively, for corn with different NIR spectrometers indicated that the former gave the best analytical accuracy, and was effective for the spectroscopic data compression which can simplify and optimize the transfer process.