用多维校正法提高近红外牛奶成分校正模型稳健性的研究

Research on the Robustness Improvement of Calibration Model for Measuring the Contents of Components in Milk by Multidimensional Calibration in Near-Infrared Spectroscopy

乳品质量问题是制约乳品工业发展的瓶颈。近红外光谱技术以其快速、简便等优越性广泛应用于各个领域,在乳品检测中有巨大的应用潜力和前景。但样品温度对近红外分析模型的稳健性和检测结果会产生较大影响,为获得较好的分析准确度,文章提出了抑制温度影响的新方法——多维正交校正(NOSC)结合多维偏最小二乘(NPLS)法。它将温度作为建模元素组成含温度信息的三维光谱矩阵,先将三维光谱阵进行预处理,运用NOSC方法滤除无关信息的干扰,再采用NPLS方法建立牛奶成分与透射光谱的校正模型。文章在25,30,35和40℃四个温度点处做了研究,波长范围为1100～1700nm。通过常规法和多维校正法来减小温度的影响。实验结果表明,常规法无法抑制温度的影响,而与其他方法相比,NOSC-NPLS法建立的校正模型具有更好的预测能力和稳健性,能更有效地消除温度的影响。

A new hybrid algorithm （NOSC-NPLS）, the combination of multi-way orthogonal signal correction （N-OSC） algorithm and multi-way partial least squares （N-PLS） algorithm, is proposed. In NOSC-NPLS algorithm, the 3-D spectral matrix was firstly constructed, which is composed of the temperature information and the NIR spectrum. Secondly, the N-OSC algorithm was used as a pretreatment algorithm to remove the interference information irrelevant to analyte in 3-D spectral matrix. Finally, the N-PLS algorithm was applied to develop the calibration model based on the pretreated 3-D transmission spectral matrix and content matrix of major components in milk. In order to evaluate the performances of conventional algorithms and multidimensional calibration algorithms on suppressing the effects of temperature variation, a batch of milk samples at temperature of 25, 30, 35 and 40℃ were measured in the wavelength range from 1 100 to 1 700 nm and the experimental results were investigated. It was found that the conventional algorithms, which could not suppress the effects caused by temperature variation, failed to obtain satisfactory results. However, compared with these algorithms, the experimental results showed that the NOSC-NPLS algorithm can effectively eliminate the effects of temperature variation and also can help achieve the analytic models with better prediction ability and robustness.