Bee Research Institute, Chinese Academy of Agricultural Sciences, Beijing 100093, China.
Food Chem. 2012 Nov 15;135(2):338-42. doi: 10.1016/j.foodchem.2012.02.156. Epub 2012 Mar 3.
The feasibility of near infrared (NIR) spectroscopy and multivariate analysis as tools to classify Chinese honey samples according to their different floral origins was explored. Five kinds of honey, namely, acacia, linden, rape, vitex and jujube, were analysed using a NIR spectrophotometer with a fibre optic probe. Classification models based on the NIR spectra were developed using Mahalanobis-distance discriminant analysis (MD-DA) and a back propagation artificial neural network (BP-ANN). By the MD-DA model, total correct classification rates of 87.4% and 85.3% were observed for the calibration and validation samples, respectively, while the ANN model resulted in total correct classification rates of 90.9% and 89.3% for the calibration and validation sets, respectively. By ANN, the respective correct classification rates of linden, acacia, vitex, rape and jujube were 97.1%, 94.3%, 80.0%, 97.1%, and 85.7% in calibration, and 100%, 93.3%, 80.0%, 100%, and 73.3% in validation. The results indicated that NIR combined with a classification technique could be a suitable technology for the classification of Chinese honeys from different botanical origins.
探索了近红外（NIR）光谱和多元分析作为根据不同花卉来源对中国蜂蜜样品进行分类的工具的可行性。使用带有光纤探头的 NIR 分光光度计分析了五种蜂蜜，分别为刺槐、椴树、油菜、荆条和大枣。基于 NIR 光谱建立了马氏距离判别分析（MD-DA）和反向传播人工神经网络（BP-ANN）分类模型。通过 MD-DA 模型，校准和验证样品的总正确分类率分别为 87.4%和 85.3%，而 ANN 模型的校准和验证集的总正确分类率分别为 90.9%和 89.3%。通过 ANN，在验证集中，椴树、刺槐、荆条、油菜和大枣的正确分类率分别为 97.1%、94.3%、80.0%、97.1%和 85.7%，在验证集中，分别为 100%、93.3%、80.0%、100%和 73.3%。结果表明，NIR 结合分类技术可能是一种适合于对来自不同植物来源的中国蜂蜜进行分类的技术。
