基于近红外透射光谱的乳制品蛋白质、脂肪含量检测

Determination of the contents of protein and fat in dairy product based on near infrared transmittance spectroscopy

应用近红外透射光谱(NITS)法对乳制品中蛋白质和脂肪含量进行快速检测。首先分别对光谱进行二阶导数加S-G平滑(SD+S-G)和一阶导数加多元散射校正加S-G平滑(FD+MSC+S-G)预处理;然后对处理后的光谱进行小波基为db3、分解尺度为6的小波压缩;最后以压缩后光谱数据作为输入变量,采用径向基函数人工神经网络(RBF-ANN)建立4种乳制品的蛋白质和脂肪定量分析模型。经过反复实验得出最佳扩散常数spread值,其中,蛋白质模型在spread值为135时预测精度最高,其相关系数(R)和预测集均方差(RMSEP)分别为0.999 9和0.030 1,脂肪模型在spread值为105时预测精度最高,其R和RHSEP分别为0.999 7和0.096 8。结果表明,基于RBF-ANN和小波压缩建模更稳定、精度更高,可以实现乳制品品质快速无损检测。

The contents of protein and fat in dairy products are determined quickly by the near infrared transmittance spectroscopy （NITS）. The speetroscopies are preprocessed by the two mixed methods：the second derivative adding S-G smoothing （SD＋S-G） and the first derivative adding muhiplicative scatter correction adding S-G smoothing （FD＋ MSC＋ S-G）, respectively, and thenthe processed spectroscopic data are compressed by the wavelet with function db3 and compression level 6. The quantitative analysis models of protein and fat in the four dairy products are established by radial basis function artificial neural network （RBF-ANN） using the compressed spectroscopy data as the input variables. The best spread value is obtained by repeated experiment. When the spread is 135, the prediction accuracy of protein is the highest and the correlation coefficient and mean square error are 0. 999 9 and 0. 0301 ,respectively. In the same way, when the spread is 105, the prediction accuracy of fat is the highest and the correlation coefficient and mean square error are 0. 999 7 and 0. 096 8, respectively. The results show that the model based on RBF-ANN combined with wavelet is more stable and with a higher accuracy. It could be used to test the qualities of dairy products quickly and destructively.