摘要
Thenear-infrared(NIR)diffusereflectancespectroscopywasusedtostudythecontentofBerberineintheprocessedCoptis.TheallocatedproportionsofCoptistoginger,yellowliquororEvodiarutaecarpachangedaccordingtotheresultsoforthogonaldesignaswellasthetemperature.Foraswithdrawingthefullandeffectiveinformationfromthespectraldataaspossible,thespectraldatawaspreprocessedthroughfirstderivativeandmultiplicativescattercorrection(MSC)accordingtotheoptimizationresultsofdifferentpreprocessingmethods.Firstly,themodelwasestablishedbypartialleastsquares(PLS);thecoefficientofdetermination(R2)ofthepredictionwas0.839,therootmeansquarederrorofprediction(RMSEP)was0.1422,andthemeanrelativeerror(RME)was0.0276.Secondly,forreducingthedimensionandremovingnoise,thespectralvariableswerehighlyeffectivelycompressedviathewavelettransformation(WT)technologyandtheHaarwaveletwasselectedtodecomposethespectralsignals.AfterthewaveletcoefficientsfromWTwereinputintotheartificialneuralnetwork(ANN)insteadofthespectrasignal,thequantitativeanalysismodelofBerberineinprocessedCoptiswasestablished.TheR2ofthemodelwas0.9153,theRMSEPwas0.0444,andtheRMEwas0.0091.Thevaluesofappraisalindex,namelyR2,RMSECV,andRME,indicatethatthegeneralizationabilityandpredictionprecisionofANNaresuperiortothoseofPLS.TheoverallresultsshowthatNIRspectroscopycombinedwithANNcanbeefficientlyutilizedfortherapidandaccurateanalysisofroutinechemicalcompositionsinCoptis.Accordingly,theresultcanprovidetechnicalsupportforthefurtheranalysisofBerberineandothercomponentsinprocessedCoptis.Simultaneously,theresearchcanalsoofferthefoundationofquantitativeanalysisofotherNIRapplication.
出版日期
2008年06月16日（中国期刊网平台首次上网日期，不代表论文的发表时间）
