以兆赫波偵測葉菜中硝酸鹽之研究

蔬菜是人類飲食當中重要的項目之一，但蔬菜常因栽種方式不當或天氣因素影響而殘留過量可轉化為致癌物的硝酸鹽。兆赫波（Terahertz wave）為一20世紀80年代中期才興起的電磁波研究領域，具有許多獨特且極具發展潛力的性質。本研究目的即為測試兆赫波是否可作為檢測蔬菜中硝酸鹽含量的有效方法，若兆赫波能檢測出硝酸鹽，則進一步明確探討相關的檢測指標性質。本研究分別使用京都大學提供傅立葉轉換紅外光譜儀之FARIS-1與浙江大學提供之時域型兆赫波光譜儀TAS7500SP對樣本進行檢測，兩者之間的差異是可檢測的頻率範圍不同。使用傅立葉轉換紅外光譜儀FARIS-1進行的檢測中，以衰減式全反射（ATR）方式檢測NaNO3、KNO3、NH4NO3與Ba(NO3)2四種硝酸鹽粉末、萵苣菜粉、青江菜葉與硝酸鈉溶液，以透射方式檢測硝酸鹽與萵苣菜粉製成的pellets（錠片）與青江菜葉，並觀察分析各檢測結果。檢測結果顯示，本研究檢測的硝酸鹽在兆赫波範圍內具有吸收特徵，但萵苣菜粉沒有吸收特徵，且當不同物質混合時，混合物的兆赫波吸收譜並非成分物質各自吸收譜的疊加。在使用時域型兆赫波光譜儀TAS7500SP進行檢測時，以衰減式全反射方式檢測45個不同硝酸鹽濃度的青江菜葉片，以透射方式檢測相同的青江菜葉片與3組不同硝酸鹽濃度的萵苣菜粉製成的共93個pellets，並將檢測結果通過不同的濃度分組方式與4種不同數據前處理方法建立176組SIMCA分類模型，通過比較其分類準確度來辦別以兆赫波檢測蔬菜中硝酸鹽含量的最佳指標與參數。在以SIMCA模型分析測得的數據並進行比較後，發現在將樣本分組時，組別之間濃度差異較大的話可以更精確地分類。比較ATR測葉片、透射測葉片與透射測pellets三種檢測組合的結果，發現最佳的數據前處理方式為使用Autoscale＋Savitzky–Golay平滑化或僅使用Autoscale。上述三種組合中，以透射測pellets的分類效果最佳，以測得的樣本相位移（Phase Shift）或折射率（Refractive Index）性質作為指標，可以建立校正正確率與驗證正確率均為100%的模型，但缺點是製作pellets為破壞性檢測。透射測葉片結果稍差，但以測得的樣本對數透射度（Transmittance(dB)）性質作為指標時，也可建立校正正確率為100%與驗證正確率為80%的模型，且其優勢是其為非破壞性檢測。ATR測葉片在三種方式中的分類效果最差，但以相位移（Phase Shift）作為指標時，仍可建立校正正確率為100%與驗證正確率為60%的模型，且也屬於非破壞性檢測。雖然以非破壞性方式檢測葉片的結果較破壞性檢測稍差，但由於其仍有一定程度的分類正確率，故顯示兆赫波技術具有以非破壞性方式檢測葉菜中硝酸鹽含量的潛力。本研究結果發現，兆赫波具有檢測蔬菜中硝酸鹽的潛力，並且以樣本相位移（Phase Shift）與葉菜中硝酸鹽含量最為相關：這顯示兆赫波極可能成為一個能有效地檢測蔬菜中硝酸鹽含量的技術。然而，由於目前兆赫波仍有許多性質未被解明，故未來仍必須對兆赫波之形成原因及基本性質進行更多研究，才能將兆赫波技術進一步實用化。

關鍵字

兆赫波；硝酸鹽；葉菜；衰減式全反射；相位移

並列摘要

Vegetables are important foods. High amount of nitrates in human body would cause health problems; and one of the sources of the nitrates in human body is from eating vegetables. Terahertz wave is a new research aspect of spectrum field. Terahertz has many unique and potential characteristics, but it had not been intensively studied until mid-1980s. The purpose of this research is to find out whether terahertz wave could be used as an effective method to detect the nitrate in vegetables. If it proves that terahertz wave can show the features of nitrates, the next step is to figure out which property of the samples could be the best property for detection. In this research, the Fourier transform infrared spectrometer (FTIR) FARIS-1 and terahertz time-domain spectrometer (THz-TDS) TAS7500SP were provided by Kyoto University and Zhejiang University, separately. The difference between these two devices is different detecting frequency range. Regarding the experiments using the FTIR FARIS-1 device, 4 kinds of nitrate powder (NaNO3、KNO3、NH4NO3 and Ba(NO3)2), Romaine lettuce powder, Bok Choy leaves and NaNO3 solution were measured in the attenuated total reflectance (ATR) mode; while the pellets made by nitrate powder and Romaine lettuce powder and Bok Choy leaves were measured in the transmission mode. The results showed that the nitrate samples have absorption features in the terahertz frequency range, while the Romaine lettuce powder has no feature in the range. Furthermore, it was found when using terahertz wave to measure the mixture of two materials, the result spectrum is not the addition of the two individual spectra of the materials. As of the experiments using the THz-TDS TAS7500SP device, 45 Bok Choy leaf samples with different nitrate contents were measured in both ATR and transmission modes. In addition, 93 pellets were measured in transmission mode, which were made from 3 groups of Romaine lettuce powder with different nitrate contents. The detecting results were combined with 2 grouping methods and 4 data preprocessing methods to build 176 SIMCA classification analyses. The best property of using terahertz wave to discriminate vegetables with different nitrate contents can be found by comparing the correction rate of classification in each analysis. Comparing the results of the SIMCA analyses, it was found the grouping method of keeping a nitrate content gap between the groups can lead to a better precise classification. Comparing the SIMCA results of using ATR to measure leaves, using transmission to measure leaves and using transmission to measure pellets, the best data preprocessing method was found to be using autoscale and Savitzky–Golay smoothing at the same time or using autoscale alone. Using transmission to measure pellets is the best pattern for using terahertz wave to separate vegetables samples with different nitrate contents. By choosing phase shift or refractive index as the target property, the transmission and pellets combination pattern can lead to an almost perfect SIMCA model, whose calibration correct rate and validation correct rate are both 100%. The only disadvantage of this pattern is making pellets must destruct the origin sample. The transmission and leaves combination pattern can build a model whose calibration correct rate is 100% and validation correct rate is 80% when choosing Transmittance (dB) as target property, while the ATR and leaves combination pattern can build a model whose calibration correct rate is 100% and validation correct rate is 60% when choosing phase shift as target property. It shows that terahertz wave still has the potential to be a non-destructive detection method to detect the nitrate content in leafy vegetables although the non-destructive detection result was not as good as the destructive detection result in the pellet form. This research proved that terahertz certainly has the potential to detect the nitrate in vegetables. The most important finding is that phase shift has very good relationship with the nitrate content in leafy vegetables. However, many basic properties of terahertz wave still remain unclear nowadays. More research on terahertz wave is needed to advance the technology toward applications.