可見光/近紅外光光譜儀用於還原乳與不同廠牌鮮奶之檢測

Abstract

由於鮮奶與還原奶之價格差異大，故有廠商以奶粉調製成還原奶，當鮮奶銷售謀利。為顧及消費者的權益，有必要發展快速鑑別還原奶及鮮奶之方法。本文利用可見光/近紅外光分光光度計掃瞄噴霧乾燥台大牧場鮮奶產生之奶粉(88%)及水(12%)所調配之還原奶與鮮奶樣本，在40℃下之光譜吸收值，利用5種不同的回歸與分類方法MLR(Multiple Linear Regression;)、PLSR(Partial Least Squares Regression)、KNN(K Nearest Neighbors )、SVM( Support Vector Machine )、及Classify( Matlab所內建的程式)、分辨還原奶、台大鮮奶、及不同廠牌(I牌與U牌)之鮮奶。5種不同的回歸與分類方法對台大還原奶與鮮奶之原始光譜值、減去水之光譜值、一階差分光譜值、平滑化處理光譜值、蛋白質光譜值、及油之光譜值之平均鑑別率依序為92%、94.5%、83.5、98.5%、94.5%、及94.5%。6種光譜值對5種不同的回歸與分類方法之平均鑑別率依序為92.5%、96.25%、84.2%、98.3%、及93.3%。U牌與I牌鮮奶用Stepwisefit挑選特徵波長之光譜值，最高平均鑑別率96%，選用馬氏距離，Stepwisefit，蛋白質，及油的特徵波長，Classify分析方法最高平均鑑別率97.5%。

Due to the price difference between reconstituted milk and fresh milk, some companies mixed powder milk with water to sell as fresh milk for making profit. To maintain consumer rights, it is necessary to develop a rapid method for classifying reconstituted milk and fresh milk. Using visible/ near infrared spectroscopy,88% spray-dry ing powder milk from the experimental farm of National Taiwan University (NTU)mixed with 12% water to make reconstituted milk and fresh milk at 40℃ were scanned , and analyzed by MLR(Multiple Linear Regression;)、PLSR(Partial Least Squares Regression)、KNN(K Nearest Neighbors )、SVM( Support Vector Machine )、及Classify( Matlab) for classifying reconstituted milk, NTU, U brand, and I brand fresh milk. Using five different regression and classification to original spectra, subtracting water spectra first difference spectra, smooth spectra ,protein spectra and oil spectra, the average classification rates in order were 92%、94.5%、83.5、98.5%、94.5%、and 94.5% respectively .Using five different regression and classification for six different spectra, the average classification rates in order were92.5%、96.25%、84.2%、98.3%、及93.3% respectively。Using the spectra corresponding to selected wavelength by Mahalanobis distance ,Stepwisefit, protein and oil, the highest average classification rate between U brand and I brand fresh milk was 97.5%。