关键词: 松江大米；矿物元素指纹；判别分析；产地溯源

Abstract: The contents of 40 mineral elements in 180 Songjiang and non-Songjiang rice samples were determined by inductively coupled plasma mass spectrometry (ICP-MS). The mineral elements fingerprints were characterized by multivariate statistics including analysis of variance (ANOVA), correlation analysis, principal component analysis (PCA) and discriminant analysis, and a discriminant model for geographical origin traceability of Songjiang and non-Songjiang rice samples was developed. The results showed that the samples from different geographical origins displayed significantly different mineral profiles with significant correlations among mineral elements. Sixteen of these 40 mineral elements were selected for dimensionality reduction. As a result, the first five PCA factors were obtained. The geographical origins of rice samples were accurately discriminated from each other by a score plot of the first three PCA factors. The discriminant model based on the eight characteristic mineral elements B, Na, Fe, Co, Ni, Zn, As and Se constructed by linear discriminant analysis gave an overall correct discrimination rate of 93.0% to the training set. Besides, the discrimination accuracy was 95.2% (sensitivity) for Songjiang rice samples and 86.8% (specificity) for non-Songjiang rice samples. The cross-validation accuracy of discrimination was 84.2%–95.2% in the training set. The overall discrimination accuracy, sensitivity and specificity for the validation set was 92.1%, 96.0% and 84.6% in back substitution test, respectively. The overall discrimination accuracy for the validation set agreed with that for the test set, proving that the established model had good discrimination accuracy, sensitivity, specificity and stability and therefore, it was feasible to discriminate Songjiang rice from non-Songjiang rice.

Key words: Songjiang rice; mineral elements; characteristic indicators; geographical origin traceability