頁籤選單縮合
題 名 | 稻米近紅外線光譜之多變數分析=The Multivariate Analyses of Near-Infrared Spectra of Rice |
---|---|
作 者 | 劉昌群; 彭智平; 蕭介宗; | 書刊名 | 農業機械學刊 |
卷 期 | 6:3 1997.09[民86.09] |
頁 次 | 頁9-19 |
分類號 | 434.113 |
關鍵詞 | 近紅外線; 多重線性回歸; 主成分迴歸; 部份最小平方迴歸; Near-infrared; NIR; Multiple linear regression; MLR; Principal component regression; PCR; Partial least squares regression; PLSR; |
語 文 | 中文(Chinese) |
中文摘要 | 多重線性迴歸(Multiple linear regression; MLR)在過去廣泛被使用於近紅外線光譜 資料的分析上,近來又發展出主成分迴歸(Principal component regression; PCR)以及部份最 小平方迴歸(Partial least squares regression; PLSR)使得在近紅外線(NIR)光譜資料的分析 的方法上有更多的選擇。 這三種光譜資料分析的方法各有其理論基礎,校正線在預測的精確度上也可能不同。本 研究的目的在於使用完全相同的原始光譜資料以上述三種不同的方法來分析,並比較其結果 以作為爾後使用近紅外線分光光度計於水稻化學成分的非破壞性檢驗上之參考。 研究的結果顯示:使用相同的樣本來建立校正線,PCR與PLSR預測性能要比以MLR 來得好。利用統計方法來剔除額外的樣本所建立之校正線,呈現PLSR優於PCR優於MLR 的趨勢,並有顯著差異。均勻的樣本之預測結果也比隨機樣本好。 不論使用PLSR、PCR或MLR原始光譜所建立的校正線,若預測性能已經十分良好, 再以一階或二階差分處理之光譜來建立校正線,其預測性能無法顯著提高。 |
英文摘要 | Multiple linear regression (MLR) has been used extensively in the past to develop near- infrared (NIR) spectrum calibrations. Today, principal componect regression (PCR) and partial least squares regression (PLSR) become popular methods chosen for calibration development. Recent literature shows that above mentioned methods have their respective theory background. The prediction of calibration curve may result in different accuracy. The objectives of this study are to analyze the identical original spectra by the three methods, to compare these results, and to dtermine which method is the best for future non-destructive detecting agricultural chemical compositions by a near infrared spectrophotometer. Results showed the same samples were used to build calibration curves which have better prediction by PCR and PLSR models than that of MLR model. By choosing suitable samples for each model, the prediction showed PLSR better than PCR and PCR better than MLR with significant difference. Besides, the prediction is better by choosing uniform samples than that by random. By PLSR、PCR or MLR, the calibration curves from the spectra with first and second difference method can't improve that prediction significantly if the prediction is good enough from original spectra. |
本系統中英文摘要資訊取自各篇刊載內容。