开展了独立成分分析(independent component analysis,ICA)联合支持向量机(support vector machine,SVM)模型对食品掺伪的判别.通过傅里叶变换红外光谱仪获得食用植物油(正品油、掺伪油和炸货油)以及奶粉(纯奶粉和掺三聚氰胺奶粉)的中红外光谱,首先,探究不同的光谱预处理方法对所建立模型的影响,经考察选取归一化结合移动平滑对光谱进行预处理;然后采用特征矩阵联合近似对角化(joint ap-proximate diagonalization of eigenmatrices,JADE)方法提取光谱特征信息,其结果优于主成分分析对特征信息的提取效果;通过Kennard-Stone算法划分训练集和测试集,利用贝叶斯优化(Bayesian optimization,BO)对SVM模型参数进行优化.经考察,所构建的JADE-BO-SVM模型对掺伪食品的识别准确度达到100%,该法可为食品掺伪的高效、准确鉴别提供新的途径和思路.
JADE-SVM mid-infrared food adulteration discrimination model based on Bayesian optimization
Independent component analysis(ICA)combined with support vector machine(SVM)model was carried out to identify food adulteration.Mid-infrared spectra of edible vegetable oil(certified oil,adulterate oil and fried oil)as well as milk powder(pure milk powder and melamine-containing milk powder)were obtained by Fourier transform infrared spectroscopy(FTIR).Firstly,the effects of different spectral preprocessing methods on the established model were investigated,and the normalization combined with the moving smoothing method was selected for the preprocessing of the spectra.Then the joint approximate diagonalization of eigenmatrices(JADE)method was used to extract the spectral feature information,and the results were better than those of the principal component analysis for feature information extraction.The training and testing sets were divided by the Kennard-Stone algorithm,and Bayesian optimization(BO)was used to optimize the SVM model parameters.Upon examination,the constructed JADE-BO-SVM model achieved 100%accuracy in identifying adulterated food.The proposed method can provide new way and new idea for the efficient and accurate identification of food adulteration.