基于低场核磁共振技术的鲜牛奶冷藏天数的鉴别

Identification of fresh milk cold storage days based on low field nuclear magnetic resonance technique

应用低场核磁共振技术结合简单分类算法(SIMCA)、线性判别法(LDA)和支持向量机法(SVM)对不同冷藏天数的鲜牛奶进行鉴别,并比较了SIMCA、LDA中不同函数及SVM中不同类型参数、不同核函数的建模效果。结果表明:LDA中Mahalanobis函数建立的模型优于Linear、Quadratic函数的模型;SVM中C-SVM类型的模型优于Nu-SVM类型的模型,径向基函数与线性函数的模型优于S型函数、多项式函数的模型。SIMCA模型的总识别准确率为95.83%,LDA中Mahalanobis函数建立的模型总识别准确率为100%,SVM中C-SVM类型的径向基函数建立的模型总识别准确率为87.50%。由此表明LDA中用Mahalanobis函数建立的模型最适合预测鲜牛奶的冷藏天数。

The study was based on the application of the low field nuclear magnetic resonance（ NMR） technology combined with simple classification algorithm（ SIMCA）,linear discriminant method（ LDA） and support vector machine method（ SVM） to identify the milk in different days cold storage,and also to compare with the modeling effect of SIMCA and LDA in different function and SVM in the arguments of different types,different kernel functions. The results showed that the building function model of the LDA in Mahalanobis was better than the function model of Linear and Quadratic,the model of C- SVM in SVM was better than that model of Nu- SVM,the function model of the radial basis and linear was better than the function of S- shaped and polynomial. The total recognition accuracy of SIMCA model was 95.83%,the total recognition accuracy of the LDA Mahalanobis building function model was 100%,the total recognition accuracy was 87.50%,which was based on the radial basis function of C- SVM in SVM Mahalanobis model.The results showed that the using of Mahalanobis model in LDA was the most suitable for predicting the cold days of fresh milk.