三维荧光光谱结合PCA-SVM对几种浓香型白酒的鉴别

提出一种利用三维荧光光谱技术鉴别不同品牌浓香型白酒的方法。运用FLS920荧光光谱仪测量了七个不同品牌浓香型白酒的三维荧光光谱，不同品牌浓香型白酒的荧光光谱特征相似，仅凭荧光特征参数较难区分。采用求偏导和小波压缩相结合的数据预处理方法，求解光谱数据中每一激发波长下，荧光强度对发射波长的一阶和二阶偏导数，选取db7紧支撑正交小波对数据进行压缩，选择4尺度分解后的近似系数作为新的数据矩阵，然后做主成分分析(PCA)。将提取的主成分作为支持向量机(SVM)的输入，并利用K-fold交叉验证的方法寻找支持向量机的最优参数c和γ，建立不同品牌白酒的分类鉴别模型。从每个品牌白酒中随机选取14个样本，共98个样本组成训练集，其余的42个样本组成预测集。分别比较了数据不求偏导，对数据求一阶偏导和二阶偏导的预处理后对鉴别模型的影响。结果表明：三维荧光光谱经过二阶偏导的预处理后，结合主成分分析和支持向量机能很好地实现不同品牌浓香型白酒的分类鉴别，模型的准确率为98.98%，预测集的准确率为100%。该方法具有简单，快速，成本低的优点，可为中国白酒的检测和鉴别技术的发展提供帮助。

The Discrimination of Chinese Strong Aroma Type Liquors with Three-Dimensional Fluorescence Spectroscopy Combined with Principal Component Analysis and Support Vector Machine

In this paper ,a method for discrimination of different bands liquor with strong aroma type based on three‐dimensional fluorescence spectrum technology was developed .Firstly ,the three‐dimensional fluorescence spectra of seven different brands liquor were measured by the FLS920 fluorescence spectrometer which produced by Edinburgh in England .The spectral shows that different bands liquors have similar fluorescence characteristics and it ’s difficult to distinguish them only with Fluorescent characteristic parameters .Because of this ,the first‐order and second‐order partial derivatives respect to fluorescence emission wavelength on each of the excitation wavelength were carried out in this paper .Daubechies‐7 (db7) orthonormal wavelet with compact support was used to compress the spectral data .The forth approximate coefficients were finally chosen as the new data matrix .Then the new data matrix was analyzed by principal component analysis (PCA) and the principal components were ex‐tracted to be used as the inputs of support vector machine (SVM ) .The K‐fold cross validation was applied to optimize the pa‐rameters c and γand the prediction model was constructed in the end .Fourteen samples were selected randomly from each brand that in total of ninety‐eight samples were selected as the training set ,and the rest forty‐two samples were collected as the predic‐tion set .The effect of three different spectral data after processing on the model is compared :original data ,the first‐order and second‐order partial derivatives on the spectral data .The results show that the three‐dimensional fluorescence spectra with the pretreatment of second‐order partial derivatives coupled with PCA and SVM can make a good performance on the brands identifi‐cation of strong aroma type liquors ,the accuracy of the established model and prediction accuracy were 98.98% and 100% ,re‐spectively .This method has the advantage of easy operation ,high speed ,low cost and provides a good help in the detection and identification of Chinese liquor .