近红外反射光谱(NIRS)技术分析奶粉品质的研究

奶粉中蛋白质和脂肪是影响奶粉营养品质的主要因素,利用近红外反射光谱分析技术对来自国内不同地区的奶粉共900份样品进行蛋白质和脂肪成分测定分析。研究了不同的样品数日、光谱预处理和散射校正技术对发展奶粉近红外测定定标模型的影响。结果表明．在样品数目为200—400范围内建立的定标分析模型较理想;数学预处理中以一阶导数较好,且以“1,4,4,1”的处理组合最为理想;光谱散射校正中采用“标准正态变量转换(SNV) 趋势变换法(De—trending)”的组合建立回归方程效果较好。利用改进最小二乘法回归技术(Modified PLS)建立多种定标模型,并进行交叉验证(cross—Validation)来分析各种因素对定标模型的影响。同时筛选出较理想的蛋白质和脂肪定标分析回归方程,其中蛋白质和脂肪含量的相关系数高速0．973和0．850。探讨了NIRS技术在建模应用中的一些影响因素,以及由NIRS技术建立奶粉分析模型用于快速分析和在线检测的可行性。

Studies on Quality Analysis of Milk Powder by Near-Infrared Reflectance Spectroscopy (NIRS)

The protein and fat are main nutrients directly affecting the quality of milk powder. In this project, protein and fat content analysis of 900 samples, gathered in different areas of the country, were carried out for developing better calibration equations for these nutrients. Meanwhile, main aspects, such as the sample number for calibration, pre-treatment of spectrum, scatter correct technology were also conducted to optimize calibration equations. The result shows that the calibration model established with sample size of 200 to 400 is more ideal. The mathematic treatment by the way of first derivative proved a better approach, and the treatment of "1,4,4,1" is suitable for calibration regression. In scatter correction, it is more ideal to establish regression models by using correction method of SNV and De-trending. Variant calibration models are established by applying regression method of modified PLS, which were tested with external cross-validation, to analyze the effects of above aspects on calibration model. Based on all the tests, ideal regression models of protein and fat were established with the correlation coefficient as high as 0.973 and 0.850, respectively. Factors influencing calibration and the possibility of NIRS technology applied to rapid predicting ingredient of milk powder and monitoring on-line are also discussed.