近红外光谱分析中的化学计量学算法研究新进展

Progress of Chemometric Algorithms in Near-infrared Spectroscopic Analysis

近红外光谱是一种绿色、快捷的分析技术,在科学研究、工业生产以及日常检测中得到广泛应用。化学计量学算法的应用在近红外光谱技术的发展过程中发挥了重要作用。化学计量学方法通过寻找测量变量之间的相关性,构建数学模型,量化样本间的差异性,并发现事物变化的内在规律,实现较合理准确的未知预测。这也是“大数据”战略的重要环节和主旨所在。该文针对近红外光谱吸收信号较弱、谱峰重叠严重,以及光谱测量过程中易受背景、噪声、无信息变量和外界环境因素干扰等,导致借助化学计量学方法建立的光谱与研究目标的定性定量分析模型变差问题,总结了近年来在近红外光谱领域所提出的一些化学计量学新方法,包括光谱预处理、变量选择、多元校正和模型转移,从不同角度阐述了这些方法在消除近红外光谱模型的干扰因素,提高模型的可靠性、预测准确性和适用性等方面的作用。

Near-infrared(NIR)spectroscopy is a green,fast analytical technology,and thus has been widely used in scientific research,industrial production and routine detection.The application of chemometric algorithms plays an important role in the development of NIR spectroscopy.Chemometrics focuses on exploring the relation between the measured variables,modeling the differences among samples in a qualitative or quantitative way,finding out the underlying trend of intrinsic sample changes,and predicting unknown samples reasonably and accurately.This is also the thumb of the“big data”strategy.This review discusses the issues commonly encountered in NIR spectroscopy,concerning the weakness of spectral signals,the serious overlapping of NIR bands,the interference from background,noise,non-informative variables or environmental factors,etc.,which could either mislead to an incorrect qualitative or quantitative analysis model relating the NIR spectroscopic measurements to target compositions of samples or worsen the model in terms of prediction capacity and accuracy.Furthermore,it also describes new chemometric methods with respect to spectral preprocessing,variable selection,multivariate calibration and calibration transfer.These methods have been proposed or developed in recent years to improve the reliability,accuracy and applicability of the chemometric NIR spectral models.