为探讨基于可见-近红外光谱技术快速检测牛奶中是否含有三聚氰胺的可行性。 文章通过往液态奶中添加不同含量的三聚氰胺, 共制备样本160个。 利用Handheld Field Spec光谱仪获取样本光谱, 其后采用不同的预处理方法对光谱数据进行预处理, 然后分别建立数学模型, 比较模型的好坏, 得到采用移动平均平滑作为数据的预处理方法较好。 从160个样本中随机的取出120个样本建模, 剩下的40个样本作为独立的验证集。 采用偏最小二乘回归法（PLS）和最小二乘支持向量机法（LS-SVM）方法分别建立判别分析模型, 利用独立的验证集对判别模型进行了预测验证。 预测结果的预测相关系数（R2）分别为0.917 4(PLS）和0.910 9(LS-SVM）, 预测标准误差（RMSEP）分别为0.030 4(PLS）和0.046 7(LS-SVM）。 研究结果表明近红外反射光谱可以作为一种快速检测牛奶中三聚氰胺的方法。

In order to investigate the feasibility of near infrared reflectance spectroscopy (NIRS) method for detecting if milk was adulterated with melamine or not, the present work has done the following research. Through adulterating different content of melamine into pure milk, altogether 160 samples were prepared. Using the Handheld Field Spec spectrometer spectral data of the samples were obtained, followed by different pretreatment methods to carry on processing the spectrum data, then establishing the mathematical model separately through comparison with different calibration models using different pretreatment methods, thus we got smoothing of moving average as the pretreatment method. One hundred twenty samples were taken out randomly from 160 samples (all) to set model, with the remaining 40 samples as the validation samples. Two discriminant analysis models were developed by using partial least squares (PLS) method and least squares-support vector machine (LS-SVM) method respectively, and then the other 40 samples were used to test the performance of the models. The coefficients of correlation (r) between the real values and the discriminant analysis models predicted ones were 0.917 4 (PLS) and 0.910 9 (LS-SVM). The root mean standard errors of prediction (RMSEP) were 0.030 4 (PLS) and 0.046 7 (LS-SVM). The results of this study indicated that NIRS method could provide rapid determination for melamine in milk.