光谱学与光谱分析
光譜學與光譜分析
광보학여광보분석
SPECTROSCOPY AND SPECTRAL ANALYSIS
2014年
2期
381-384
,共4页
车蜡%Vis-NIR光谱%线性判别方法%最小二乘支持向量机%连续投影算法
車蠟%Vis-NIR光譜%線性判彆方法%最小二乘支持嚮量機%連續投影算法
차사%Vis-NIR광보%선성판별방법%최소이승지지향량궤%련속투영산법
Car wax%Vis-NIR spectroscopy%Linear discrimination analysis (LDA)%Least-square support vector machine (LS-SVM )%Successive projections algorithm (SPA )
探讨了可见-近红外光谱技术快速无损识别不同品牌车蜡的可行性。实验一共获得104样本,其中40个样本(建模集)用于建立模型,剩余64个样本(预测集)被用于独立验证建立好的模型。基于五种不同品牌车蜡的可见-近红外光谱分别建立了线性判别分析(linear Discriminant Analysis ,LDA )和最小二乘支持向量机(least square-support vector machine ,LS-SVM )模型。基于两个算法的全波段光谱模型的预测集正确率分别达到了84%和97%。进一步采用连续投影算法(successive projections algorithm ,SPA)算法从751波段中选取了7个特征波段(351,365,401,441,605,926和980 nm)。基于SPA选择的变量建立LS-SVM 模型,准确率依然保持在97%。说明SPA选择的特征波段包含了对于车蜡品牌鉴别最重要的光谱信息,而大多数无用信息则被有效剔除。将SPA与LS-SVM算法的车蜡识别模型在保证正确率的基础上,还可以大大降低模型计算复杂程度,说明该模型能快速准确的从车蜡可见-近红外光谱中提取有效信息,并实现车蜡品牌的无损鉴别。
探討瞭可見-近紅外光譜技術快速無損識彆不同品牌車蠟的可行性。實驗一共穫得104樣本,其中40箇樣本(建模集)用于建立模型,剩餘64箇樣本(預測集)被用于獨立驗證建立好的模型。基于五種不同品牌車蠟的可見-近紅外光譜分彆建立瞭線性判彆分析(linear Discriminant Analysis ,LDA )和最小二乘支持嚮量機(least square-support vector machine ,LS-SVM )模型。基于兩箇算法的全波段光譜模型的預測集正確率分彆達到瞭84%和97%。進一步採用連續投影算法(successive projections algorithm ,SPA)算法從751波段中選取瞭7箇特徵波段(351,365,401,441,605,926和980 nm)。基于SPA選擇的變量建立LS-SVM 模型,準確率依然保持在97%。說明SPA選擇的特徵波段包含瞭對于車蠟品牌鑒彆最重要的光譜信息,而大多數無用信息則被有效剔除。將SPA與LS-SVM算法的車蠟識彆模型在保證正確率的基礎上,還可以大大降低模型計算複雜程度,說明該模型能快速準確的從車蠟可見-近紅外光譜中提取有效信息,併實現車蠟品牌的無損鑒彆。
탐토료가견-근홍외광보기술쾌속무손식별불동품패차사적가행성。실험일공획득104양본,기중40개양본(건모집)용우건립모형,잉여64개양본(예측집)피용우독립험증건립호적모형。기우오충불동품패차사적가견-근홍외광보분별건립료선성판별분석(linear Discriminant Analysis ,LDA )화최소이승지지향량궤(least square-support vector machine ,LS-SVM )모형。기우량개산법적전파단광보모형적예측집정학솔분별체도료84%화97%。진일보채용련속투영산법(successive projections algorithm ,SPA)산법종751파단중선취료7개특정파단(351,365,401,441,605,926화980 nm)。기우SPA선택적변량건립LS-SVM 모형,준학솔의연보지재97%。설명SPA선택적특정파단포함료대우차사품패감별최중요적광보신식,이대다수무용신식칙피유효척제。장SPA여LS-SVM산법적차사식별모형재보증정학솔적기출상,환가이대대강저모형계산복잡정도,설명해모형능쾌속준학적종차사가견-근홍외광보중제취유효신식,병실현차사품패적무손감별。
Visible and near-infrared (Vis-NIR) spectroscopy was applied to identify brands of car wax .A total of 104 samples were obtained for the analysis ,in which 40 samples (calibration set) were used for model calibration ,and the remaining 64 sam-ples (prediction set) were used to validate the calibrated model independently .Linear discriminant analysis (LDA ) and least square-support vector machine (LS-SVM ) were respectively used to establish identification models for car wax with five brands based on their Vis-NIR spectra .Correct rates for prediction sample set were 84% and 97% for LDA and LS-SVM models ,re-spectively .Spectral variable selection was further conducted by successive projections algorithm ,(SPA) ,resulting in seven fea-ture variables (351 ,365 ,401 ,441 ,605 ,926 ,and 980 nm) selected from full range spectra that had 751 variables .The new LS-SVM model established using the feature variables selected by SPA also had the correct rate of 97% ,showing that the select-ed variables had the most important information for brand identification ,while other variables with no useful information were eliminated efficiently .The use of SPA and LS-SVM could not only obtain a high correct identification rate ,but also simplify the model calibration and calculation .SPA-LS-SVM model could extract the useful information from the Vis-NIR spectra of car wax rapidly and accurately for the non-destructive brand identification of car wax .