多波长激光诱导荧光麦卢卡蜂蜜掺杂分类识别

Classification and Recognition of Adulterated Manuka Honey by Multi-Wavelength Laser-Induced Fluorescence

麦卢卡蜂蜜产自新西兰,具有很强的抗菌及抗氧化作用,其售价较高,近年来掺假事件时有发生,利用激光诱导荧光技术对掺杂糖浆的麦卢卡蜂蜜进行分类识别研究。选用266,355,405和450 nm四种常用激光作为激发源,选择三种品牌的新西兰进口麦卢卡蜂蜜(编号A,B,C)中掺杂烘焙糖浆作为实验样品,掺杂比例为0%~90%,间隔10%;每个激发波长下每种样本溶液重复测试60次,共7200组数据。光谱数据首先进行荧光波段截取、平滑及归一化等预处理;然后随机选取80%的数据做训练集,20%的数据做测试集;对训练集数据使用主成分分析(PCA)结合线性判别分析(LDA)做数据降维;最后对降维后的数据分别建立K最近邻(KNN)和支持向量机(SVM)分类模型,对测试集数据进行分类识别,重复进行50次随机分组及分类识别后对得到的分类识别率求平均值及标准差。实验分析结果表明,激发光波长对最终识别结果影响较大,266 nm激发的荧光光谱分类识别正确率最高,三种麦卢卡蜂蜜掺杂溶液的分类识别率均能达到98.5%以上,最高能达100%;355和405 nm激发的分类识别效果次之,所有样品的分类识别率均大于92%;而450 nm激发的荧光光谱分类识别率最低,不同样品的分类识别率均低于66%。因此,后续分类算法的比较仅使用266,355和405 nm激发的荧光光谱数据,分析结果表明,KNN算法的分类效果要优于SVM算法,不同激发波长下三种蜂蜜掺杂溶液的分类识别率均是KNN算法更高,且对266 nm激发的三种蜂蜜掺杂样品,KNN算法的分类识别率比SVM算法要高1%以上。实验结果表明,使用激光诱导荧光技术对掺假麦卢卡蜂蜜进行分类识别是可行的,对于掺杂糖浆的麦卢卡蜂蜜,在使用的所有组合中,266 nm激发结合PCA-LDA降维和KNN分类算法的分类识别率最高,分类效果最好,可用于掺假麦卢卡蜂蜜的快速准确鉴别。

Manuka honey is produced in New Zealand and has strong antibacterial and antioxidant effects.The price is relatively high,and adulteration incidents have occurred frequently in recent years.This paper uses laser-induced fluorescence(LIF)technology to classify and identify Manuka honey adulterated with syrup.Four commonly used lasers of 266,355,405 and 450 nm are selected as excitation sources,and three brands of New Zealand Manuka honey(No.A,B and C)adulterated with baking syrup are used as experimental samples.The adulteration ratio ranged from 0%to 90%,with an interval of 10%.Each sample solution has been tested 60 times under different excitation wavelengths,with a total of 7200 sets of data.For the spectral data,firstly,pretreat with fluorescence band interception,smoothing and normalization;Then,randomly select 80%of the data as training and 20%as test sets.Principal component analysis(PCA)combined with linear discriminant analysis(LDA)is used to reduce the dimension of the training set data;Finally,K-nearest neighbor(KNN)and support vector machine(SVM)classification models are established for the dimensionality reduction data respectively,and the test set data are classified and identified by the models.After 50 times of random grouping and classification calculation,the recognition rate’s average value and standard deviation are obtained.The experimental results show that the excitation wavelength greatly influences the final recognition results.The recognition rate of 266 nm excitation is the highest.The recognition rates of the three Manuka adulterated solutions are more than 98.5%,and the highest can reach 100%;355 and 405 nm excitation are the second,and the recognition rates of all samples are greater than 92%;The classification effect of 450 nm excitation is the worst,with the recognition rates less than 66%.Therefore,the comparison of classification algorithms only uses the spectral data excited by 266,355 and 405 nm.The analysis results show that the classification effect of the KNN algorithm is better than the SVM algorithm.For the three honey adulterated solutions excited by 266 nm,the recognition rates of the KNN algorithm are more than 1%higher than that of the SVM algorithm.According to the experimental results,using LIF to classify and identify adulterated Manuka honey is feasible.For Manuka honey adulterated with syrup,among all combinations used in this paper,266 nm excitation combined with PCA-LDA and KNN algorithms has the highest recognition rate and the best classification effect,which provides an effective method for rapid and accurate identification of adulterated Manuka honey.