拉曼光谱基线校正

2Spectroscopy29(2)February2014(1,2).TheRamanspectrashowthecharacteristicsforspeciesinsharpanddensepeaks.However,duringtheapplicationofRamanspectroscopy,fluorescenceoforganiccompoundsinthesamples,whicharesometimesseveralordersofmagnitudemoreintensethantheweakRamanscatter,caninterferewiththeRamansignals(3).Aphenomenonofbaselinedriftshowsup,makingtheresolutionandanalysisofRamanspectraimpractical.Bothinstrumental(4)andmathematicalmethodshavebeendevelopedtoreducethedriftedbaselinecausedbyfluorescence.Theuseofanexcitationwavelengthsuchas785–1064nmlasers,whichdoesnoteliminatefluorescence(5),isthemosttraditionalinstrumentalmethod.Ramanscatteringisdirectlyproportionaltothefourthpoweroffrequency;astheexcitationwavelengthincreases,thesen-sitivityoftheRamanbecomesseverelyreduced.Theuseofanti-StokesRamanspectroscopyisanothermethod,basedontheory(6).Mathematicalmethods(7–10)includethefirstandsecondorderderivatives,wavelettransform,me-dianfilter,andmanualpolynomialfitting.Thesemethodsareusefulincertainsituations,butstillhavesomelimita-tions.Forexample,derivativesareeffective,butasaresulttheshapeoftheRamanspectrumischanged;wavelettrans-formcanbedifferentiableinthehigh-andlow-frequencycomponentsofthesignals;however,itisdifficulttochooseadecompositionmethod.Manualpolynomialfittingsre-quiretheusertoidentifythe“non-Raman”locationsmanu-ally(11),andafterwardsthebaselinecurveisformedbyfittingtheselocations.Consequently,theresultinvolvestheinevitablesubjectivefactorsand,inaddition,theworkloadisalwaysheavy.Therefore,itisimportanttochooseanop-timaldecompositionmethod.Piecewiselinearfittingbasedoncritical-point-seekingwasproposedinthisstudy.Themethoddeterminesanop-timumcorrectedspectrumbycorrelationanalysis,whichcanconquertheselimitations.ARamanspectrumfromthesulfamicacidcatalyticreactionofanaspirinsystemwasusedasastudysubject.Byusingthismethod,theRamanspectrumdriftedbaselinewasautomaticallyeliminated,leavingonlythecorrectedspectrum.TheoryandMethodBasisofQualitativeandQuantitativeRamanAnalysisARamanspectrumisaplotoftheintensityofRamanscatteredradiationasafunctionofitsfrequencydiffer-encefromtheincidentradiation(usuallyinunitsofwave-numbers,cm-1).ThisdifferenceiscalledtheRamanshift,whichisthebasisofqualitativeanalysis(12).TheintensityorpowerofanormalRamanpeakdependsinacomplexwayuponthepolarizabilityofthemolecule,theintensityKuoSun,HuiSu,ZhixiangYao,andPeixianHuangThecorrectionofbaselinedriftisanimportpartfordatapreprocessing.Anintervallinearfittingmethodbasedonautomaticcritical-point-seekingwasimproved,whichmadeitpossibleforthebaselinetodriftautomatically.Experimentaldatawereacquiredfromthesulfamicacidcatalyticreactionoftheaspirinsystem,whichconsistedofdifferentproportionsofaspirin.Asimulatedbase-linewithdifferentintervalvaluesofmovingaveragesmoothingdeterminedsettingparametersinthismethod.Afterbaselinedriftscausedbyfluorescenceareremoved,thedifferencesofcharacter-isticaspirinpeaksprovedtheefficiencyofthismethod.BaselineCorrectionforRamanSpectraBasedonPiecewiseLinearFittingFebruary2014Spectroscopy29(2)3(13,14).Equation1showsthefactorsthatdeterminetheRamanscatteringcrosssection:I=K(v)×A(v)×v4×I0×J(v)×C[1]whereIistheintensityoftheRamanline;K(v)istheoverallspectrometerresponse;A(v)isself-absorptionofthemedium;visthefrequencyofscatteredradiation;I0istheintensityoftheinci-dentradiation;J(v)isamolarscatteringparameter;andCistheconcentrationofthesample.Thev4termdominatesiftheothertermsdonotdifferappre-ciably,andahigherfrequencylaserbeamyieldsastrongerRamansignal.MethodStep1:SmoothingWeusedsmoothingtopreprocessthedataforRamanspectroscopywithfrequencyshiftsandhighsensitivity.Movingaveragesmoothingcaneffec-tivelylowerthefrequencyandsensi-tivityasshowninequation2:()smooth121mk,xk,xkix=+++=–==im[2]whereωistheintervalnumberofthemovingaveragesmoothingwindow,whichmustbeanoddnumber.Step2:LocatingLocalExtremaSupposethatthefunctionf(x)hasanextremevalueatpointx=x0inacertainneighborhood(x0–δ,x0+δ)wherethederivativeofthefunctionisdefinedandisnot0.Ifxϵ(x0–δ,x0)thederivativeispositive,whereasifxϵ(x0+δ,x0)thederivativeisnega-tive,f(x0)ismaximum,otherwisef(x0)isminimum.Afterfindingthelocalminimum,wecangetasetofminimalcriticalpointsλi(i=1,2,...,n).TheRamanspectrumwasdividedinton-1intervalsbyλi.Step3:FittingBaselineEveryinterval(λ1,λ2),(λ2,λ3),...,(λi–1,λi)canbefittedinalinearequa-tionasshowninequation3:φ(x)=(f(λi)–f(λi-1))(f(λi)+f(λi-1))(x–x∈)+(λi+λi-1)(λi-1,λi)i=(2,3,...,n)(λi–λi-1)22φ(x)=(f(λi)–f(λi-1))(f(λi)+f(λi-1))(x–x∈)+(λi+λi-1)(λi-1,λi)i=(2,3,...,n)(λi–λi-1)22[3]whereφ(x)isthefittingbaseli