- 期刊
- OpenAccess
近紅外光與核磁共振應用於乳粉脂肪含量之檢測
Application of Near Infrared Spectroscopy and Nuclear Magnetic Resonance in Determination of Fat Content in Milk Powder
摘要
本研究應用近紅外光與核磁共振兩種非破壞性檢測技術，成功地建立乳粉中脂肪含量之檢測模式。近紅外光檢測部份，採用多重線性迴歸(MLR)與部份最小平方迴歸(PLSR)兩種檢測模式探討不同的光譜處理（原始光譜、一次微分光譜、二次微分光譜）對乳粉脂肪含量檢測的影響。實驗所使用之乳粉樣本數爲79個，脂肪含量範圍爲0.8～28％，波長範圍爲700～2498 nm，間隔爲2 nm。在MLR模式中，校正方程式是以一次微分光譜之五波長組合（742 nm、948 nm、1314 nm、1790 nm、2352 nm）爲最佳（r(下標 c^2)=0.9987，SEC=0.263；r(下標 p^2)=0.997，SEP=0.425，RPD=19.04）。在PLSR模式中，也是以一次微分光譜所建立的校正方程式爲最好（波長範圍700～2498nm），其校正與預測結果爲r(下標 c^2)=0.9999，SEC=0.066；r(下標 p^2)=0.997，SEP=0.455，RPD=17.79。由於PLSR模式是將整段光譜納入迴歸計算，所以其結果在各光譜處理中，均較MLR模式爲佳。本研究之結果顯示，適當的光譜處理，如光譜平滑化、區段取點等，可以有效地消除雜訊幹擾所產生的誤差，提高校正與預測的準確性。核磁共振部份，分爲自由感應衰退(FID)檢測與自旋迴訊(Spin Echo)檢測。所得結果以自由感應衰退的檢測結果較好，其校正與預測結果爲r(下標 c^2)=0.9805，SEC=1.027；r(下標 p^2)=0.990，SEP=0.820，RPD=9.87。近紅外光與核磁共振兩種檢測方法，皆有快速且非破壞的共同優點。但就檢測精準度而言，近紅外光因爲利用多變數的迴歸統計方法(MLR、PLSR)，並且可以挑選較不受乳粉中其他成份（如水份）影響的波長或波段，所以近紅外光的檢測結果較核磁共振好。
並列摘要
Non-destructive inspection techniques including near infrared (NIR) spectroscopy and nuclear magnetic resonance (NMR) were successfully used to determine the fat content in milk powder in this study. In near infrared spectroscopy research, multiple linear regression (MLR) and partial least square regression (PLSR) were used to develop models for three kinds of spectrum treatments (original, first derivative and second derivative). The range of the spectra was from 700 to 2498 nm with a wavelength increment of 2 nm. Seventy-nine milk powder samples (fat content range 0.8～28%) were used for analysis in this study. In MLR analysis, the best equation of calibration was obtained by using first derivative spectra with five wavelengths (742 nm, 948 nm, 1314 nm, 1790 nm and 2352 nm); the results of calibration and prediction were r(subscript c^2)=0.9987, SEC=0.263; r(subscript p^2)=0.997, SEP=0.425, RPD=19.04. In PLSR analysis, the best equation of calibration was also obtained by using first derivation spectra (700～2498 nm); the results of calibration and prediction were r(subscript c^2)=0.9999, SEC=0.066; r(subscript p^2)=0.997, SEP=0.455, RPD=17.79. Since all the wavelengths in the specified range (700～2498 nm) were included in PLSR analysis, the results of PLSR analysis in original, first derivative and second derivative spectra were all better than those in MLR analysis. As shown in this study, appropriate spectrum treatments, such as smoothing and gap techniques could eliminate the spectrum noises effectively, and gave better results in calibration and prediction. In nuclear magnetic resonance research, the response of hydrogen nuclei was measured by Free Induction Decay (FID) and Spin Echo. Regarding the determination of fat content in milk powder, the performance of FID was better than that of Spin Echo. The FID results of calibration and prediction were r(subscript c^2)=0.9805, SEC=1.027; r(subscript p^2)=0.990, SEP=0.820, RPD=9.87. Non-destructive and rapid measurement was the common advantage for both near infrared spectroscopy and nuclear magnetic resonance methods. However, regarding the accuracy, near infrared spectroscopy is better than nuclear magnetic resonance, because near infrared spectroscopy uses multi-variable statistics (MLR and PLSR) and the wavelengths could be such selected that other constituents such as water in milk powder would not have influence, therefore, near infrared spectroscopy method is better than nuclear magnetic resonance method in determining the fat content in milk powder.