- 期刊
Using Orthogonal Signal Correction to Correct the Near Infrared Reflectance Spectroscopy for Determination of Protein Content in Brown Rice
利用直交訊息校正法修正近紅外線光譜以測定糙米之蛋白質含量
摘要
在進行糙米蛋白質含量的測定時，常將糙米粒磨成粉狀後，再利用近紅外光譜（nearinfrared reflectance spectroscopy; NIRS）建立檢量模式。然而，NIRS容易受到不同組成的干擾影響。為了有效降低在使用NIRS時的干擾影響，本研究以normalization、standard normal variate（SNV）、multiplicative signal correction（MSC）及兩種直交訊息校正法（orthogonal signalcorrection; OSC），包括OSC-CV及JSosc 等方法，對經由糙米粒之NIRS資料進行前處理，再利用淨最小平方回歸（partial leastsquares regression; PLSR）建立檢量模式，以進行蛋白質含量的預測。研究結果顯示，經過OSC-CV前處理方法轉換後，在均方根誤差及R2等統計指標的表現上，均較其他的前處理方法為佳。此外，OSC-CV修正糙米粒之NIRS後建立的檢量模式，其分析能力與利用糙米粉之NIRS所獲得的結果相近。並且，利用OSC-CV修正糙米粒之NIRS後建立的檢量模式時，其潛在變數之數目較糙米粉之NIRS所建立的檢量模式為少，因而使得所建立的模式更容易解釋。
並列摘要
When determining the protein content of brown rice (unpolished grain of rice), one usually grinds the rice grains into flour and then adopts the near infrared reflectance spectroscopy (NIRS) to construct the calibration model. However, the NIRS is easily influenced by interference among different constituents. To effectively eliminate interference in NIRS on brown rice, this study employed several data pre-treatment methods, including normalization, standard normal variate (SNV), multiplicative signal correction (MSC), and two orthogonal signal correction (OSC) approaches, which were OSC-CV and JSosc. Then, the partial least squares regression (PLSR) was used to construct a calibration model to determine the protein content of brown rice. Results showed that the PLSR model constructed by using OSC-CV to correct the NIRS of brown rice had the better statistical indicators in terms of the root mean squared error and coefficient of determination (R2) than the other pre-treatment methods. The performance of the PLSR model was similar to those constructed on the original NIRS of the brown rice flour. Moreover, the number of latent variables required in the PLSR model was less than that with the original NIRS of brown rice flour. In conclusion, the OSC-CV method was a useful tool for correction and interpretation when constructing the PLSR model for determining the protein content of brown rice with NIRS data.