简介:Severalfeaturesofretinalvesselscanbeusedtomonitortheprogressionofdiseases.Changesinvascularstructures,forexample,vesselcaliber,branchingangle,andtortuosity,areportentsofmanydiseasessuchasdiabeticretinopathyandarterialhypertension.Thispaperproposesanautomaticretinalvesselsegmentationmethodbasedonmorphologicalclosingandmulti-scalelinedetection.First,anilluminationcorrectionisperformedonthegreenbandretinalimage.Next,themorphologicalclosingandsubtractionprocessingareappliedtoobtainthecruderetinalvesselimage.Then,themulti-scalelinedetectionisusedtofinethevesselimage.Finally,thebinaryvasculatureisextractedbytheOtsualgorithm.Inthispaper,forimprovingthedrawbacksofmulti-scalelinedetection,onlythelinedetectorsat4scalesareused.Theexperimentalresultsshowthattheaccuracyis0.939forDRIVE(digitalretinalimagesforvesselextraction)retinaldatabase,whichismuchbetterthanothermethods.
简介:Inordertoslovethelarge-scalenonlinearprogramming(NLP)problemsefficiently,anefficientoptimizationalgorithmbasedonreducedsequentialquadraticprogramming(rSQP)andautomaticdifferentiation(AD)ispresentedinthispaper.Withthecharacteristicsofsparseness,relativelylowdegreesoffreedomandequalityconstraintsutilized,thenonlinearprogrammingproblemissolvedbyimprovedrSQPsolver.Inthesolvingprocess,ADtechnologyisusedtoobtainaccurategradientinformation.Thenumericalresultsshowthatthecombinedalgorithm,whichissuitableforlarge-scaleprocessoptimizationproblems,cancalculatemoreefficientlythanrSQPitself.
简介:ThispaperpresentstwodifferentalgorithmsthatderivethecohesionstructureintheformoflexicalchainsfromtwokindsoflanguageresourcesHowNetandTongYiCiCiLin.There-searchthatconnectsthecohesionstructureofatexttothederivationofitssummaryisdisplayed.Anovelmodelofautomatictextsummarizationisdevised,basedonthedataprovidedbylexicalchainsfromoriginaltexts.Moreover,theconstructionrulesoflexicalchainsaremodifiedaccord-ingtocharacteristicsoftheknowledgedatabaseinordertobemoresuitableforChinesesumma-rization.EvaluationresultsshowthathighqualityindicativesummariesareproducedfromChi-nesetexts.
简介:ThepaperpresentsanalgorithmofautomatictargetdetectioninSyntheticApertureRadar(SAR)imagesbasedonMaximumAPosteriori(MAP).Thealgorithmisdividedintothreesteps.First,itemploysGaussianmixturedistributiontoapproximateandestimatemulti-modalhistogramofSARimage.Then,basedontheprincipleofMAP,whenaprioriprobabilityisbothunknownandlearnedrespectively,thesamplepixelsareclassifiedintodifferentclassesc={target,shadow,background}.Last,itcomparestheresultsoftwodifferenttargetdetections.Simulationresultspreferablyindicatethatthepresentedalgorithmisfastandrobust,withthelearnedaprioriprobability,anapproachtotargetdetectionisreliableandpromising.
简介:Beingdirectedagainettwokinksofnoiseinopticalfibersensors,asim-pleandeffectivemethodofautomaticcompensationforopticalfibersensorsispresented.Notonlytheunstabilityeffectoflightsource,butalsozerodriftofphotoelectronicde-vices,canbeeliminatedorenormouslyrestranedwiththeaidofthismethod.Inanotherway,byusingsingle-chipmrcrocomputer,theopticalfibersensorsystemfabricatediscaonnetedtoacopmuternetworkstorealizeanautomaticcompensation.
简介:ThepaperaddressestheproblemoftargetrecognitionusingHigh-resolutionRadarRangeProfiles(HRRP).Anovelapproachoffeatureextractionanddimensionreductionbasedonextendedhighordercentralmomentsisproposedinordertoreducethedimensionofrangeprofiles.FeaturesextractedfromradarHRRPsarenormalizedandsmoothed,andthencomparativeanalysisofthesimilarapproachesisdone.Therangeprofilesareobtainedbystepfrequencytechniqueusingthetwo-dimensionalbackscattersdistributiondataoffourdifferentaircraftmodels.Thetemplatematchingmethodbynearestneighborrules,whichisbasedonthetheoryofkernelmethodsforpatternanalysis,isusedtoclassifyandidentifytherangeprofilesfromfourdifferentaircrafts.Numericalsimulationresultsshowthattheproposedapproachcanachievegoodperformanceofstability,shiftindependenceandhigherrecognitionrate.Itishelpfulforreal-timeidentificationandtheengineeringimplementsofautomatictargetrecognitionusingHRRP.Thenumberofrequiredtemplatescouldbereducedcon-siderablywhilemaintaininganequivalentrecognitionrate.
简介:UseSingle-chipComputertoRealizeAutomaticCompensationforOpticalFiberSensors①WANGShouyu,TIANGuangyun,YINGuiliang(YanshanUnivers...
简介:Theneedforwide-bandclockanddatarecovery(CDR)circuitsisdiscussed.A2Gbpsto12Gbpscontinuous-rateCDRcircuitemployingamulti-modevoltage-controloscillator(VCO),afrequencydetector,andaphasedetector(FD&PD)isdescribed.Anewautomaticfrequencybandselection(FBS)withoutexternalreferenceclockisproposedtoselecttheappropriatemodeandalsosolvetheinstabilityproblemwhenthecircuitispoweringon.Themulti-modeVCOandFD/PDcircuitswhichcanoperateatfull-rateandhalf-ratemodesfacilitateCDRwithsixoperationmodes.TheproposedCDRstructurehasbeenmodeledwithMATLABandthesimulatedresultsvalidateitsfeasibility.
简介:ThispaperproposesaPCAandKPCAself-fusionbasedMSTARSARautomatictargetrecognitionalgorithm.Thisalgorithmcombinesthelinearfeatureextractedfromprincipalcomponentanalysis(PCA)andnonlinearfeatureextractedfromkernelprincipalcomponentanalysis(KPCA)respectively,andthenutilizestheadaptivefeaturefusionalgorithmwhichisbasedontheweightedmaximummargincriterion(WMMC)tofusethefeaturesinordertoachievebetterperformance.Thelinearregressionclassifierisusedintheexperiments.Theexperimentalresultsindicatethattheproposedself-fusionalgorithmachieveshigherrecognitionratecomparedwiththetraditionalPCAandKPCAfeaturefusionalgorithms.
简介:AmeasurementsystemwiththeCCDmatrixandcomputersystemisdesignedtotestthe2Dsizeofanyshapeworkpiecesautomatically.Inaddition,thesystemadoptsthemethodoftherelativemeasurementwhichincreasestheprecisionandthevelocity.Moreimportantly,theprecisioncan'tbechangedwiththeconditionsofthetemperatureandairpressure.Theexperimentsshowthattherelativeprecisionof0.0029andtheabsoluteprecisionof2.97μmareobtained.Theinstrumentmaybeusedintheproductlineandmakethetestingonlinepossible.
简介:Thisletterintroducesa4thorderactiveRCcomplexfilterwith1.5MHzcenterfrequencyand1MHzbandwidth.Thetotalharmonicdistortionofthefilterislessthan–60dBandtheimagerejectionratioisgreaterthan60dB.Anoveltechniqueisalsoproposedinthislettertoautomaticallyadjustthevariationofthetimeconstant.Theadvantagesoftheproposedmethodareitshighprecisionandsimplicity.Using5bitscontrolwords,thetuningerrorislessthan±1.6%.