氢核磁共振结合支持向量机鉴别蜂蜜植物源

Identification of the botanic source of honey by 1H nuclear magnetic resonance and support vector machine

中文关键词:  氢核磁共振  支持向量机  蜂蜜  植物源  主成分

英文关键词:1H nuclear magnetic resonance  support vector machine  honey  botanic source  principal component

中文摘要:

      目的 建立基于氢核磁共振(1H nuclear magnetic resonance, 1H NMR)结合支持向量机分类模型鉴别蜂蜜植物源的方法。方法 采集荆条蜜、油菜蜜、洋槐蜜、葵花蜜4种不同植物源的蜂蜜共计122例样品的谱图信息, 分全谱(δ 0.10~δ 9.50)、脂肪区(δ 0.10~δ 3.00)、糖类化合物区(δ 3.00~δ 6.00)、芳香区(δ 6.00~ δ 9.50)4个不同积分区间建立分类模型, 结合主成分权值系数筛选特征变量, 进一步优化判别模型。结果 基于主成分权值系数筛选变量范围δ 3.40~δ 3.90和δ 4.60~δ 4.70内共计267个积分变量, 以该区域积分变量为输入变量建立的支持向量机分类模型, 对训练集的判别正确率为97.53%, 对测试集的判别正确率为100%。结论 通过主成分权值系数能有效筛选特征变量, 减少输入变量的同时提高模型稳健性与准确性, 基于氢核磁共振结合支持向量机分类模型能有效鉴别不同植物源蜂蜜。

英文摘要:

      Objective To establish a method for identification of the botanic source of honey by 1H nuclear magnetic resonance (1H NMR) spectroscopy and support vector machine. Methods Spectral information of 122 samples were collected including 4 kinds of honey such as vitex honey samples, rape honey samples, acacia honey samples and sunflower honey samples. Classification models were established based on 4 different integration intervals including full spectrum (δ 0.10?δ 9.50), fat zone (δ 0.10?δ 3.00), carbohydrate zone(δ 3.00?δ 6.00) and aromatic zone (δ 6.00?δ 9.50), and the discriminant model was further optimized by screening feature variables with principal component weight coefficient. Results Based on the principal component weight coefficient, 267 integral variables were screened in the range of variables δ 3.40?δ 3.90 and δ 4.60?δ 4.70. The classification model of support vector machine was established with the regional integral variable as input variable, the discriminant accuracy of the training set was 97.53%, and the discriminant accuracy of the test set was 100%. Conclusion The weight coefficients of principal components can effectively pick the characteristic variables, reduce the input variables and improve the robustness and accuracy of the model. The classification model based on 1H NMR combines with support vector machine can identify honey from different plants effectively.