茶葉醱酵度之高光譜影像檢測

Abstract

本研究開發了實驗室型高光譜影像系統，並以此系統進行茶葉醱酵度檢測分析。系統硬體部份包括影像擷取室、影像擷取系統、液晶可調控濾鏡系統三部份，軟體方面則自行開發高光譜影像系統控制程式以及高光譜影像資訊運算程式。影像擷取室包括穩定、堅固之影像擷取室和具有待測光譜全波段響應的鹵素光源，成功建立具有高光強度、高光均勻度之光譜影像量測平面。液晶可調控濾鏡與影像擷取系統搭配自行設計之高光譜影像擷取流程，可在2.5分鐘內完成波長範圍在400-1100 nm高光譜影像之擷取，並且搭配最佳電子快門迭代運算流程以及光平面灰階校正，可確保高光譜影像具有高灰階解析度以及正確灰階響應。高光譜影像資訊運算程式使用空間校正、背景分離以及光譜擷取等步驟，成功擷取出高光譜影像中具有分析化學成份能力之光譜資訊，並且利用修正型部份最小平方迴歸(modified partial least linear regression, MPLSR)以及多重線性迴歸(multiple linear regression, MLR)等迴歸方法建立與茶葉醱酵度具有高相關性之總兒茶素類含量檢量模式，並且分析其特徵波段及特徵波長。 本研究之實驗對象為台茶12號，在2008年夏茶總兒茶素類含量的MPLSR檢量模式分析部份，其校正組檢測結果之rc可達0.99，SEC可達5.76 mg/g，而驗證組結果之rv可達0.99，SEV為6.30 mg/g；在MLR檢量模式的分析部份，其校正組檢量結果之rc可達0.99，SEC為6.49 mg/g，而驗證組結果之rv可達0.98，SEV為9.56 mg/g。在2009年春茶總兒茶素類含量的MPLSR檢量模式分析部份，校正組檢測結果之rc可達0.99，SEC可達4.56 mg/g，而驗證組結果之rv可達0.98，SEV為5.67 mg/g；在MLR檢量模式部份，校正組檢量結果之rc可達0.99，SEC為5.04 mg/g，而驗證組結果之rv可達0.98，SEV為6.50 mg/g。將檢量模式代入完成影像處理後之高光譜影像當中，搭配茶葉醱酵度定義可計算求得茶葉樣本醱酵度影像，相較於其他僅以近紅外光光譜技術檢測茶葉總兒茶素類含量之研究，由於本研究所採用之高光譜影像技術不僅可取得茶葉醱酵度之空間分佈資訊，更重要的是本研究可於製茶流程中提供製茶者茶葉醱酵度變化之資訊。 利用本研究建立之高光譜影像系統並且搭配MPLSR及MLR可成功建立茶葉總兒茶素含量之檢量模式。將檢量線代入高光譜影像後搭配茶葉醱酵度定義可得到茶葉醱酵度分佈之影像空間資訊，確實達成快速且非破壞性茶葉醱酵度檢測之目標，並且對於未來建立茶葉製程之自動化生產及品質控管之相關研究提供良好之研究方向。

Hyperspectral image system (HSIS) was developed in this study to analyze the fermentation degree of tea. The hardware of the system included three parts: (1) lighting chamber, (2) image acquisition module (IAM), and (3) liquid crystal tunable filter (LCTF). The software was also developed in this study, and consisted of HSIS control program and hyperspectral image processing program. A rigid aluminum chamber and the light source with two 100W tungsten halogen lamps were used to set up the lighting chamber. Combining with IAM, LCTF, and HSIS control program, the hyperspectral image cube (400-1100 nm) was acquired automatically and quickly in 2.5 minutes. Spatial correction, background segmentation, and spectral information extraction were used to obtain good quality of hyperspectral images. Modified least partial square regression (MLPSR) and multiple linear regression (MLR) were employed to establish the calibration models for the total catechins content of the tea leaves. The characteristic wavelength ranges and wavelengths were selected from analyses. The variety of tea studied was Taiwan tea No. 12. The analysis results of MPLSR models (2008 summer tea and 2009 spring tea) of rc, SEC, rv , and SEV were 0.99-0.99, 4.56-5.79 mg/g, 0.98-0.99 and 5.67-6.30 mg/g respectively. The analysis results of MLR models (2008 summer tea and 2009 spring tea) of rc, SEC, rv , and SEV were 0.99-0.99, 5.04-6.49 mg/g, 0.98-0.98 and 6.50-9.56 mg/g respectively. The regression model was then employed in processed hyperspectral images to evaluate the fermentation degree of tea. The measurements of the total catechins content using NIR spectral information were addressed in other previous studies, but the quantitative analysis of the fermentation degree of tea using hyperspectral imaging was achieved in this research. The fermentation degree of tea as the intensity in the image not only provides the spatial information of tea fermentation, but also supplies the variation of fermentation degree of tea during the production process. From experiment results, the performance of the hyperspectral imaging system with MPLSR and MLR provided the quantitative models for the total catechins content of tea. The spatial information of the fermentation degree of tea can be calculated by employing the regression models to the hyperspectral images. The development of the hyperspectral imaging system can be used as a rapid and non-destructive method for the automation and quality monitoring during the tea production.