高光譜影像於龍膽指標成份之檢測應用

Detection of Marker Compounds of Gentiana scabra Using Hyper-Spectral Imaging

鄭宇帆(Yu-Fan Cheng)

指導教授 ： 陳世銘

國立臺灣大學/生物資源暨農學院/生物產業機電工程學研究所/碩士(2009年)

https://doi.org/10.6342/NTU.2009.10107

摘要

近年來，光譜影像技術於植株生長以及生物材料品質分析已被廣泛應用，技術也逐漸成熟。本研究以高光譜影像技術，進行龍膽組培苗及植株光譜量測，探討光環境與栽培期時間對於龍膽指標成份(龍膽苦苷與當藥苦苷)之影響，並建立指標成份含量檢測模式。
以反應曲面設計所規劃之LED光環境設定作為試驗條件確實會造成龍膽組培苗光譜資訊與指標成份含量上的差異；龍膽植株以栽培期長短作為實驗變因，經變異數分析(Analysis of Variance, ANOVA)發現指標成份含量的變異是極為顯著的，此結果有利於後續之光譜分析。
由On-Line NIRS 6500量測而得之龍膽組培苗與植株鮮葉光譜，以修正型部分最小平方迴歸(modified partial least square regression, MPLSR)光譜迴歸分析法建構龍膽組培苗指標成份含量之檢測模式，龍膽組培苗龍膽苦苷之MPLSR分析結果，校正組RC可達0.935，且RMSEC已降低至0.33 %，RSEC更是只有5.85 %，至於預測組之RP為0.826、RMSEP為0.439 %，而RSEP相當低，僅有7.75 %。龍膽組培苗之當藥苦苷之MPLSR分析結果為相關係數RC=0.935，誤差RMSEC=0.06 %，RMSEP=0.114 %。
龍膽植株指標成份龍膽苦苷以因子數為6且最佳數學前處理(2,2,2,1)時，其校正組檢測模式為最佳(RC=0.895，RMSEC=0.422 %)。至於當藥苦苷MPLSR光譜檢測模式是以因子數為3搭配數學前處理為(2,2,2,1)的情形之下檢量線為最佳(RC、RMSEC分別為0.946、0.036 %)，不管是植株或是組培苗，均成功建立指標成份含量檢量模式。
本研究自行建立之高光譜影像系統，減少了人為主觀判斷的誤差，節省實驗進行的時間，以反射率與平均灰階強度值作線性內插計算，可快速估算出龍膽植被在波長400 nm到1100 nm下的吸收光譜。
以高光譜影像擷取之光譜影像資訊，進行MPLSR分析建立的龍膽組培苗指標成份含量檢測模式，龍膽苦苷的預測情形在校正與預測能力的表現上(RC約0.8，而RMSEC與RMSEP也分別在0.6 %以下)，優於當藥苦苷的檢測模式分析結果(RC、RMSEC、RMSEP分別為0.683、0.120 %、0.126 %)，有好的校正與預測能力。龍膽植株之龍膽苦苷與當藥苦苷檢測模式RC均可達到0.9，校正誤差RMSEC更各在0.439 %及0.04 %以下，以此檢量線結果進行預測，所得之龍膽苦苷預測誤差為0.452 %，而當藥苦苷預測誤差為0.041 %。
合併組培苗及植株樣本一起分析之龍膽苦苷之高光譜影像MPLSR之分析結果指出，校正組之RC可達0.943，而RMSEC可達到0.570 %，預測組檢量線其RP可達0.916，RMSEP也可低於1 %，而至0.699 %，具優異之校正與預測能力。另外，當藥苦苷於龍膽組培苗與植株含量差異甚大而導致光譜分析結果表現差強人意，無法找出具備校正與預測能力之檢量線。整體而言，高光譜影像資訊經過適當的光譜數學前處理後，已成功建立指標成份含量之檢量線，除了有好的校正結果之外，亦具備預測指標成份含量之能力。
綜論，本研究已建立之高光譜影像系統，透過數位影像處理法分析光譜影像資訊，轉換為吸收度光譜，並應用於龍膽指標成份含量檢測，由MPLSR光譜迴歸分析模式成功地建立組培苗與植株的龍膽苦苷與當藥苦苷兩種指標成份之檢量線，能在龍膽生長過程中，進行即時且非破壞的檢測。

關鍵字

龍膽 ； 近紅外光 ； 修正型部分最小平方迴歸法 ； 光譜影像

並列摘要

Using spectral imaging technology to monitor the physiological status of plants has become feasible in recent years. In this study, we used hyper-spectral imaging technology to measure and analyze the reflectance spectra of Gentiana scabra growing in vitro and in greenhouse. The effects of light environments and growth period on the contents of marker compounds of Gentiana scabra including gentiopicroside and swertiamarin were investigated, and the calibration NIR models of these marker compounds were developed. .
Response surface method (RSM) designed LED light environments were adopted. The results indicated that both marker compounds contents and reflectance spectra were significantly influenced by the designed LED light environments using the variance analysis (the analysis of variance, ANOVA).
On-Line NIRS 6500 and modified partial least square regression (MPLSR) were used to measure and analyze the spectra of both seedlings and plants of Gentiana scabra. Regarding the seedlings, MPLSR results showed that RC = 0.935, RMSEC = 0.33% and RSEC = 5.85% of calibration set, and RP = 0.826, RMSEP = 0.439% and RSEP = 7.75% of calibration set for gentiopicroside while RC = 0.935, RMSEC = 0.06%, RMSEP = 0.114% for swertiamarin. Regarding the plants, the best calibration model for gentiopicroside, marker compounds in Gentiana scabra showed RC = 0.895, RMSEC = 0.422% while RC = 0.946, RMSEC = 0.036% for swertiamarin.
A hyper-spectral imaging system with the wavelength range of 400 – 1100 nm was developed in this study. It provided objective, nondestructive and fast spectral imaging measurements of Gentiana scabra samples. Using hyper-spectral imaging system for seedlings, it gave the results that RC was approximately 0.8, and RMSEC and RMSEP less than 0.6% for gentiopicroside; while RC, RMSEC, RMSEP were 0.683, 0.120%, and 0.126% respectively for swertiamarin. Using hyper-spectral imaging system for plants, RC, RMSEC, RMSEP were 0.9, 0.439%, and 0.452% respectively for gentiopicroside while RC, RMSEC, and RMSEP were 0.9, 0.04%, and 0.041% respectively for swertiamarin.
As for the pooled samples of seedlings and plants, the results of MPLSR yielded RC, RMSEC, RMSEP were 0.943, 0.570%, 0.699% respectively for gentiopicroside. However, the swertiamarin content in seedlings and plants were differ extremely, it seemed difficult to obtain a good calibration spectra model.
As a conclusion, a hyper-spectral imaging system was developed and was applied to the measure the contents of marker compounds in Gentiana scabra. Calibration models using MPLSR were successfully established for gentiopicroside and swertiamarin for both seedlings and plants; and it provided instant and nondestructive measurements of gentiopicroside and swertiamarin during the growth of Gentiana scabra.

並列關鍵字

Gentiana scabra ； Near Infrared (NIR) ； Modified Partial Least Square Regression (MPLSR) ； Spectral Image

參考文獻

陳金男。2005。LED光環境控制系統之建立與其應用於星辰花組培苗生理影響之探討。碩士論文。台北：國立臺灣大學生物產業機電工程學研究所。

陳加增。2007。應用智慧型光譜資訊分析於蔬菜植株氮含量檢測之研究。博士論文。台北：國立臺灣大學生物產業機電工程學研究所。

方煒、饒瑞佶。2004。高亮度發光二極體在生物產業之應用。中華農學會報5(5): 432-446。

Hsu, P. H. and Y. H. Tseng. 2004. Visualization of hyperspectral images for spectral analysis. journal of photogrammetry and remote sensing 9(3): 81-96.

蔡兆胤、謝俊夫、盛中德。2005。砷化銦鎵多通道檢波器水果檢測系統之研製。農業機械學刊14(1): 33-48。