紫外可见光谱指纹图谱结合化学计量学在鉴别咖啡不同焙炒度中的应用

Discrimination of Coffee Samples of Different Roasting Degree by UV-Vis Spectroscopic Fingerprint with the Aid of Chemometrics

采用紫外可见光谱指纹图谱结合多元数据分析建立一种可快速鉴别不同焙炒度咖啡的方法,考察不同的光谱前处理方法对样品分类结果的影响,比较不同的模式识别方法对样品的鉴别结果。结果表明:一阶导数处理被选为最优的前处理方法,大部分样品能够在主成分分析(PCA)和系统聚类分析(HCA)中按各自特性聚为一类,线性判别分析(LDA)的分类效果优于PCA和HCA;最小二乘向量机(LS-SVM)模型的预报结果优于偏最小二乘判别分析(PLS-DA)和反传人工神经网络(BP-ANN),识别率和预报率均为100%。

A rapid screening method was established to discriminate coffee of different roasting degree based on the Uhraviolet-Visible（UV-Vis） fingerprint in combination with multivariate data analysis. Different pre-processing methods were tested to evaluate the effect on sample classification, and different pattern recognition techniques were compared to choose the best one with regarding to the recognition rate and prediction rate. The results revealed that the first derivative was chosen the best pre-processing method and a well discrimination was achieved between the defined categories after performing PCA and HCA on the data matrix, the performance of LDA was better than that of PCA and HCA. LS-SVM model showed a clear improvement in the overall recognition rate（100%） and prediction rate（100%） compared with that of PLS-DA and BP-ANN.