改进偏最小二乘法在近红外牛奶成分测量中的应用

Using improved PLS methods for milk components determination by near infrared spectra

采用NicoletNexus870红外-近红外傅里叶变换光谱仪测量了36个市售巴氏杀菌纯牛乳样品的透射光谱。在近红外光谱1254～1875nm和2045～2372nm波段内,为了选择携带信息量大的波长区域,采用改进偏最小二乘回归法,包括间隔偏最小二乘法、移动窗口偏最小二乘法和可变窗宽移动窗口偏最小二乘法对巴氏杀菌纯牛乳中脂肪、蛋白质及乳糖成分分别建立模型,进行了分析和比较,结果表明,采用改进偏最小二乘法所选出的波长区与目标值的相关程度高,可以较好地建立牛奶的预测模型。

36 transmittance spectra of pasteurised milk are tested by a FT-IR-Spectrometer （Nicolet Nexus 870）. In order to find the most informative region, three improved partial-least squares regression methods, including interval partial-least square regression method （iPLS）, moving widow partial-least square regression method （MWPLSR） and changeable size moving window partial least-square regression method （CSMWPLS） are used for fat, protein and lactose model building between 1254nm to 1875nrn and 2045nm to 2372nm. The result shows that the spectra region optimized by CSMWPLSR can relate the targets better and thus build a good milk component predictive model.