近红外光谱技术结合人工神经网络鉴别生鲜奶和蛋白掺假奶

Study on Discrimination of Raw Milk and Milk Adulterated Foreign Protein Based on Near-infrared Spectroscopy and Artif Icial Neural Net Work Model

以牛奶和分别掺有羊奶、豆浆的掺假奶为原料,利用近红外光谱仪对样品进行扫描并得到光谱数据,应用主成分分析结合人工神经网络技术对试验数据进行多元统计分析,研究鉴别原料乳蛋白掺假的可行性。分析结果表明:应用主成分分析法,得到能反映牛奶99.32%光谱信息的6个主成分。由这6个主成分得到的得分图,可以区分掺羊奶和豆浆的牛奶,但不能完全区分生鲜奶;选用人工神经网络进行进一步信息提取与种类判别,将6个主成分作为人工神经网络的输入,对应的牛奶种类作为输出,建立了一个三层BP神经网络模型,模型对建模集84个样本的鉴别率为96.23%,,对预测集21个样本的鉴别率为95.24%。说明该方法能快速无损地鉴别原料乳中的蛋白掺假。

In order to find out a fast measure method of adulterated milk based on near infrared spectroscopy, raw milk and milk adulterated with goat milk and soybean milk were collected respectively. Using near-infrared spectroscopy to scan the samples and get the spectrum data. Then all data were analyzed by principal component analysis and artificial neural network. Results show that the accumulative reliabilities of the first six components was more than 99.32%. According to the first six components, the authors could distinguish milk adulterated with goat milk and soybean milk, but could not deal with all of raw milk. So the authors chose ANN-BP as further research method. The first six components were then applied as ANN-BP inputs and the values of the type of milk were applied as the outputs. A three-layer back propagation neural network model was developed for classification. Finally, the result indicated the distinguishing rate of 84 calibration samples is 96.23% and the distinguishing rate of 21 unknown test samples is 95.24%. All of these suggested that near infrared spectroscopy has good potential to qualitative detect adulterated milk rapidly and nondestructively.