拉曼散射光譜於臺灣烏龍茶葉產地判別模型之應用

Abstract

烏龍茶半發酵茶的一種，在臺灣本土茶市場中具有高經濟價值及高市佔率。臺灣烏龍茶產地眾多其價格各有不同，在高海拔地區所產之茶葉因其具有較鮮甜且清爽的口感，因此受到民眾喜愛，價格也較低海拔地區所生產之茶葉價格高。茶葉的品質優劣受到氣候、土壤和製茶程序等因素影響，為確保品質均一，合理的茶葉併堆是必要的，但如為謀取高額利潤，將低價茶葉併堆高價茶葉，則有損臺灣茶品質優良的美名，因此正確判別臺灣茶產地來源將有助於建立臺灣茶品牌誠信、穩定高品質市場及保障消費者權益。 本研究使用表面增強型拉曼(SERS)，結合多變量分析，進行茶產地、季節和海拔判別分析。分析之茶樣採收於春天與冬天，產地分別於3個縣市和4個山區–南投縣、桃園市、新北市坪林區、大禹嶺、梨山、阿里山、杉林溪，種植地區海拔高度，分為3個等級–低海拔、中高海拔和高海拔。對於茶葉化學物理性質，本研究選了14種物化參數進行量測，其中5種參數－游離胺基酸、總多酚、GCG、EGCG和GC對於產地判別具有顯著差異。 透過拉曼光譜的取得，使烏龍茶之指紋圖譜得以建立，可辨識出茶鹼、可可鹼、咖啡因、兒茶素和L茶胺酸等五項成分之特徵峰位置，各產區圖譜在特定特徵峰雖有些微差異但不明顯。茶樣拉曼圖譜藉由3種分類器–軟獨立建模分類法(SIMCA)、線性判別分析(LDA)和支持向量機(SVM)進行產地、季節和海拔之判別模型。軟獨立建模分類法(SIMCA)之準確率優於另外2種分類器，其產地、季節和海拔辨識率分別為85%、75%和80%。從此結果顯示，表面增強型拉曼(SERS)指紋圖譜可應用於識別茶種類，亦可結合多變量分析應用於臺灣烏龍茶茶產地鑑別。

Oolong tea, one of the semi-oxidized teas, is a high profitable tea types and has a large market share in Taiwan. The geographic origin of oolong tea is one of the major factors for its market price due to the special flavor of high-mountain tea, which taste sweet and fresh. Therefore, the price of tea from high-mountain is higher than from low-elevation area. Surface-enhance Raman scattering (SERS) is a novel spectroscopic method for compositional analysis, and it is selected in this study to develop classification models to identifying the locations, seasons and elevations of oolong tea. Tea samples used in this study were from seven locations: Nantou, Taoyuan, Pinglin, Dayuling, Lishan, Ali and Senlin. Besides, all of samples were collected in spring and winter, and the elevations were defined as elementary, intermediate and superior. There were 14 physicochemical parameters were measured to describe the properties of physical and chemical of tea. Among of these parameters, free amino acid, total polyphenol, GCG, ECG, GC were used to discriminate the locations that reached statistical significance. Using Raman spectroscopy, the fingerprint spectra of oolong tea was developed. The locations of five featured Raman peaks were identified including theophylline, theobromine, caffeine, catechins, and L-theanine. Slightly compositional differences on Raman spectra of different origins were observed but there is no statistical significance. Soft independent modeling of class analogy (SIMCA), linear discriminant analysis (LDA) and support vector machine (SVM) were adopted to develop the classification models with SERS spectra. SIMCA performed a better accuracy for classification than others. The accuracy for locations, seasons, and altitudes were 85%, 75%, and 80%, respectively. A predictive model was developed for identifying the geographical origins of oolong tea in Taiwan using SERS and multivariate methods.