光谱学与光谱分析
光譜學與光譜分析
광보학여광보분석
SPECTROSCOPY AND SPECTRAL ANALYSIS
2015年
6期
1539-1545
,共7页
涂斌%宋志强%郑晓%曾路路%尹成%何东平%亓培实
塗斌%宋誌彊%鄭曉%曾路路%尹成%何東平%亓培實
도빈%송지강%정효%증로로%윤성%하동평%기배실
稻米油%近红外光谱技术%定性-定量%特征波长%偏最小二乘法%支持向量机
稻米油%近紅外光譜技術%定性-定量%特徵波長%偏最小二乘法%支持嚮量機
도미유%근홍외광보기술%정성-정량%특정파장%편최소이승법%지지향량궤
Rice bran oil%near infrared spectroscopy%Qualitative-Quantitative%Feature wavelength%Partial least squares%Sup-port vector machine
该文主要研究激光近红外光谱分析技术结合化学计量学方法对稻米油掺伪进行定性‐定量分析。分别将大豆油、玉米油、菜籽油、餐饮废弃油掺入稻米油中，按照不同质量比配置189个掺伪油样，利用激光近红外光谱仪采集光谱；对采集的稻米油掺伪图谱数据进行多元散射校正（M SC ）、正交信号校正（OSC ）、标准正态变量变换和去趋势技术联用算法（SNV＿DT ）三种不同预处理并与原始数据进行比较。采用连续投影算法（SPA ）对经过预处理的光谱数据进行特征波长提取，应用支持向量机分类（SVC ）方法建立稻米油掺伪样品的定性分类校正模型，选择网格搜索算法对模型参数组合（C ， g）进行寻优，确定最优参数组合。另采用后向间隔偏最小二乘法（BiPLS）和SPA对预处理后的光谱数据进行特征波长提取，分别应用偏最小二乘法（PLS）和支持向量机回归（SVR）建立掺伪油含量的定量校正模型，并选用网格搜索算法对SVR模型参数组合（C ， g）进行寻优，建立最优参数模型。研究表明，建立的SVC模型预测集和校正集的准确率分别达到了95％和100％；对比SVR和PLS方法建立的数学模型对稻米油中掺杂油脂的含量的预测，两种方法均能够实现含量预测，SVR模型的预测能力更好，相关系数 R高于0.99，均方根误差（MSE）低于5.55×10－4，预测精度高。结果表明，采用激光近红外光谱分析技术可以实现稻米油掺伪的定性‐定量分析，同时为其他油脂的掺伪分析提供了方法。
該文主要研究激光近紅外光譜分析技術結閤化學計量學方法對稻米油摻偽進行定性‐定量分析。分彆將大豆油、玉米油、菜籽油、餐飲廢棄油摻入稻米油中，按照不同質量比配置189箇摻偽油樣，利用激光近紅外光譜儀採集光譜；對採集的稻米油摻偽圖譜數據進行多元散射校正（M SC ）、正交信號校正（OSC ）、標準正態變量變換和去趨勢技術聯用算法（SNV＿DT ）三種不同預處理併與原始數據進行比較。採用連續投影算法（SPA ）對經過預處理的光譜數據進行特徵波長提取，應用支持嚮量機分類（SVC ）方法建立稻米油摻偽樣品的定性分類校正模型，選擇網格搜索算法對模型參數組閤（C ， g）進行尋優，確定最優參數組閤。另採用後嚮間隔偏最小二乘法（BiPLS）和SPA對預處理後的光譜數據進行特徵波長提取，分彆應用偏最小二乘法（PLS）和支持嚮量機迴歸（SVR）建立摻偽油含量的定量校正模型，併選用網格搜索算法對SVR模型參數組閤（C ， g）進行尋優，建立最優參數模型。研究錶明，建立的SVC模型預測集和校正集的準確率分彆達到瞭95％和100％；對比SVR和PLS方法建立的數學模型對稻米油中摻雜油脂的含量的預測，兩種方法均能夠實現含量預測，SVR模型的預測能力更好，相關繫數 R高于0.99，均方根誤差（MSE）低于5.55×10－4，預測精度高。結果錶明，採用激光近紅外光譜分析技術可以實現稻米油摻偽的定性‐定量分析，同時為其他油脂的摻偽分析提供瞭方法。
해문주요연구격광근홍외광보분석기술결합화학계량학방법대도미유참위진행정성‐정량분석。분별장대두유、옥미유、채자유、찬음폐기유참입도미유중，안조불동질량비배치189개참위유양，이용격광근홍외광보의채집광보；대채집적도미유참위도보수거진행다원산사교정（M SC ）、정교신호교정（OSC ）、표준정태변량변환화거추세기술련용산법（SNV＿DT ）삼충불동예처리병여원시수거진행비교。채용련속투영산법（SPA ）대경과예처리적광보수거진행특정파장제취，응용지지향량궤분류（SVC ）방법건립도미유참위양품적정성분류교정모형，선택망격수색산법대모형삼수조합（C ， g）진행심우，학정최우삼수조합。령채용후향간격편최소이승법（BiPLS）화SPA대예처리후적광보수거진행특정파장제취，분별응용편최소이승법（PLS）화지지향량궤회귀（SVR）건립참위유함량적정량교정모형，병선용망격수색산법대SVR모형삼수조합（C ， g）진행심우，건립최우삼수모형。연구표명，건립적SVC모형예측집화교정집적준학솔분별체도료95％화100％；대비SVR화PLS방법건립적수학모형대도미유중참잡유지적함량적예측，량충방법균능구실현함량예측，SVR모형적예측능력경호，상관계수 R고우0.99，균방근오차（MSE）저우5.55×10－4，예측정도고。결과표명，채용격광근홍외광보분석기술가이실현도미유참위적정성‐정량분석，동시위기타유지적참위분석제공료방법。
The purpose of this study is mainly to have qualitative‐quantitative analysis on the adulteration in rice bran oil by near‐infrared spectroscopy analytical technology combined with chemo metrics methods .The author configured 189 adulterated oil samples according to the different mass ratios by selecting rice bran oil as base oil and choosing soybean oil ,corn oil ,colza oil , and waste oil of catering industry as adulterated oil .Then ,the spectral data of samples was collected by using near‐infrared spec‐trometer ,and it was pre‐processed through the following methods ,including without processing ,Multiplicative Scatter Correc‐tion(MSC) ,Orthogonal Signal Correction(OSC) ,Standard Normal Variate and Standard Normal Variate transformation De‐Trending(SNV_DT) .Furthermore ,this article extracted characteristic wavelengths of the spectral datum from the pre‐pro‐cessed date by Successive Projections Algorithm (SPA) ,established qualitatively classified calibration methods of adulterated oil through classification method of Support Vector Machine(SVM) ,optimized model parameters(C ,g) by Mesh Search Algorithm and determined the optimal process condition .In extracting characteristic wavelengths of the spectral datum from pretreatment by Backward interval Partial Least Squares (BiPLS ) and SPA ,quantitatively classified calibration models of adulterated oil through Partial Least Squares(PLS) and Support Vector Machine Regression (SVR) was established respectively .In the end , the author optimized the combination of model parameters (C ,g) by Mesh Search Algorithm and determined the optimal parame‐ter model .According to the analysis ,the accuracy of prediction set and calibration set for SVC model reached 95% and 100% re‐spectively .Compared with the prediction of the adulteration oil content of rice bran oil which was established by the PLS model , the SVR model is the better one ,although both of them could implement the content prediction .Furthermore ,the correlation coefficient R is above 0.99 and the Root Mean Square Error (MSE) is below 5.55 × 10-4 .The results show that the near‐infra‐red spectroscopy technology is effective in qualitative‐quantitative analysis on the adulteration of rice bran oil .And the method is applicable to analyze adulteration in other oils .
