目前我国蜂蜜市场掺假现象严重, 研究一种快速、 准确的方法用于市场流通领域掺假蜂蜜的鉴别具有重要的现实意义。 采用近红外光谱(NIR)结合化学计量学方法对常见的天然蜂蜜以及掺假(掺杂常见糖浆)蜂蜜进行建模识别, 并比较偏最小二乘-判别分析(PLS-DA)及支持向量机(SVM)对糖浆掺假蜂蜜鉴别模型的影响。 首先, 采集来自中国10个省份、 20种常见蜂蜜的112个天然纯蜂蜜样品, 以及6种常见糖浆样品按不同糖浆含量(10%, 20%, 30%, 40%, 50%, 60%)配制的112个掺假蜂蜜样品, 共计224个样品; 通过近红外光仪器扫描获得所有样品的近红外光谱数据(波长范围400~2 500 nm); 然后, 分别采用一阶导数(FD)、 二阶导数(SD)、 多元散射校正(MSC)、 标准正态变化(SNVT)四种方式对原始光谱进行预处理; 再结合PLS-DA和SVM建立天然蜂蜜和糖浆掺假蜂蜜的鉴别模型, 比较不同预处理方法对两种不同建模算法建立的蜂蜜掺假鉴别模型效果。 其中SVM算法的惩罚参数c和核函数参数g通过网格搜索法(GS)、 遗传算法(GA)、 粒子群算法(PSO)三种寻优算法进行优化。 分析结果表明: 光谱数据进行预处理后所建立的模型准确率均有明显提升, 而对于SVM模型, 惩罚参数c和核函数参数g对模型准确率的提升效果要比光谱预处理带来的提升效果更明显。 在PLS-DA算法中, 经FD光谱预处理后建立的模型效果最好, 最佳PLS-DA模型准确率为87.50%; 在SVM算法中, 经MSC预处理后, 再通过GS寻优, 获得惩罚参数c为3.0314, 核函数参数g为0.3298的条件下所建立的模型效果最好, 最佳SVM模型准确率为94.64%。 由此可见, 非线性的SVM算法结合NIR光谱数据所建立的天然蜂蜜与糖浆掺假蜂蜜鉴别模型要优于线性的PLS-DA模型, 同时表明NIR光谱结合化学计量学方法对常见糖浆掺杂的中国蜂蜜鉴别是可行的。

To find a fast, accurate, and effective method for the identification of honey adulteration, near-infrared spectroscopy combined with chemometrics was used to analyze natural honey and adulterated honey in this paper. First, 224 samples were collected for the study, including 112 natural pure honey samples from 20 common honeys in China, and 112 adulterated honey sample were prepared with 6 different syrup samples according to different syrup contents(10%, 20%, 30%, 40%, 50%, or 60%). Near infrared spectral data (wavelength range of 400～2 500 nm) of all samples were obtained by near infrared light instrument scanning. Then, first derivative (FD), second derivative (SD), multiple scattering correction (MSC), and standard normal variation (SNVT) pre-processing of the original spectra combined with PLS-DA (linear algorithm) and SVM (non-linear algorithm) modeling, respectively, were adopted to establish a differential model of natural honey and syrup-adulterated honey and compare the effects of different pretreatment methods on the honey adulteration identification model established by the two different modeling algorithms. The penalty parameter c and the kernel function parameter g of the SVM algorithm were optimized by three optimization algorithms: grid search, genetic algorithm, and particle swarm optimization. The analysis results showed that the PLS-DA model established by the FD preprocessing had the best effect, and the accuracy of the best PLS-DA model was 87.50%. After MSC pre-processing, the SVM model with the penalty parameter c of 3.031 4 and the kernel function parameter g of 0.329 8 was the best. The accuracy of the best SVM model was 94.64%. It can be seen that the non-linear SVM algorithm combined with the NIR spectral data natural honey and syrup-adulterated honey identification model is better than the PLS-DA model.