简介:Inthispaper,wepresentahighspeedautofocussystemformicrosystemapplicationsanddesignalook-up-tablebasedautofocusingalgorithmforapplicationswhenatargetobjectisalwaysvisible,e.g.,manufacturingpartswithalignmentfiducials.Weperformanevaluationof24focusmeasurestoverifythatwhichfocusmeasureisthebestforthelook-up-tablebasedmethod.Fromtheevaluation,wefindthattheChebyshevmoments-basedfocusmeasure(CHEB)isthemostsuitable.Furthermore,wealsodevelopalook-up-tablebasedautofocussystemthatusesCHEBasthefocusmeasure.Intrainingphase,weofflineconstructatablefromtrainingimagesofanobjectthatarecapturedatseverallensdistances.Eachentryoftableconsistsoffocusmeasurecomputedfromimageandlensdistance.Inworkingphase,givenaninputimage,thealgorithmfirstcomputesthefocusmeasureandthenfindsthebestmatchfocusmeasurefromthetableandlooksupthecorrespondinglenspositionformovingitintothein-focusposition.Ouralgorithmcanperformautofocusingwithinonly2stepsoflensmoving.Theexperimentshowsthatthesystemcanperformhighspeedautofocusingofmicroobjects.
简介:RadarSystemEngineering¥DaleR.Billetter(RaytheonCompany,Massachusetts.USA)WHATISRADARSYSTEMENGINEERING?Anorganizationdesiredto...
简介:Toolmanagementisnotasingle,simpleactivity,itiscomprisedofacomplexsetoffunctions,especiallyinaflexiblemanufacturingsystem(FMS)environment.Theissuesassociatedwithtoolmanagementincludetoolrequirementplanning,toolreal-timescheduling,toolcribmanagement,toolinventorycontrol,toolfaultdiagnosis,tooltrackingandtoolmonitoring.Inordertomaketoolsflowinto/outofFMSefficiently,thisworkisaimedtodesignaknowledge-baseddecisionsupportsystem(KBDSS)fortoolmanagementinFMS.Firstlyanoverviewoftoolmanagementfunctionsisdescribed.ThenthestructureofKBDSSfortoolmanagementandtheessentialagentsinthedesignofKBDSSarepresented.FinallytheindividualagentsofKBDSSarediscussedfordesignanddevelopment.
简介:Thispaperproposesanewneuralfuzzyinferencesystemthatmainlyconsistsoffourparts.Thefirstpartisabouthowtouseneuralnetworktoexpresstherelationwithinafuzzyrule.Thesecondpartisthesimplificationofthefirstpart,andexperimentsshowthatthesesimplificationswork.Onthecontrarytothesecondpart,thethirdpartistheenhancementofthefirstpartanditcanbeusedwhenthefirstpartcannotworkverywellinthefuzzyinferencealgorithm,whichwouldbeintroducedinthefourthpart.Finally,thefourthpart"neuralfuzzyinferencealgorithm"isbeenintroduced.Itcaninferencethenewmembershipfunctionoftheoutputbasedonpreviousfuzzyrules.Theaccuracyofthefuzzyinferencealgorithmisdependentonneuralnetworkgeneralizationability.Evenifthegeneralizationabilityoftheneuralnetworkweusedisgood,westillgetinaccurateresultssincethenewcomingrulemaynotberelatedtoanyofthepreviousrules.Experimentsshowthisalgorithmissuccessfulinsituationswhichsatisfytheseconditions.
简介:Ahybridapproachforfuzzysystemdesignbasedonclusteringandakindofneurofuzzynetworksisproposed.Anunsupervisedclusteringtechniqueisfirstlyusedtodeterminethenumberofif-thenfuzzyrulesandgenerateaninitialfuzzyrulebasefromthegiveninput-outputdata.Then,aclassofneurofuzzynetworksisconstructedanditsweightsaretunedsothattheobtainedfuzzyrulebasehasahighaccuracy.Finally,twoexamplesoffunctionapproximationproblemsaregiventoillustratetheeffectivenessoftheproposedapproach.
简介:PrologtotheSectiononRadarSystemEngineering¥//IntheEnglishlanguagewehaveanexpressionthatsaysitispossible"tolosesightofthefores...
简介:TheElectromagneticCompatibilityandDistributionofAntennaSystem¥B.FWang;S.Z.adns&Y.L.Yao(Dept.ofElect.Eng.,BeijingUniversityofA...
简介:Greysequencegenerationcandrawoutanddevelopimpliedrulesoftheoriginaldata.Differentkindsofgenerationmethodsweresummarizedandclassifiedintotwotypes:partialgenerationandwholegeneration.Theaveragegenerationandstepwiseratiogenerationisdisussed,thepreferencegenerationisregardasaspecialcaseofproportionaldivisionbasedonanalysisgeometrictheory,proposeanideaofusingconcaveandconvexstatusofdiscretedatatodeterminethegenerationcoefficient.Basedonthestepwiseandsmoothratiogeneration,atendencyaveragegenerationisproposedandhaveacomparisonusingthedataprovidedinpaperslistedinthereferences.Thecomparisonprovesthatthenewgenerationisbetterthantheothertwogenerationsanderrorsareobviouslyreduced.
简介:Theautomaticdetectionoffacesisaveryimportantproblem.Theeffectivenessofbiometricauthenticationbasedonfacemainlydependsonthemethodusedtolocatethefaceintheimage.Thispaperpresentsahybridsystemforfacesdetectioninunconstrainedcasesinwhichtheillumination,pose,occlusion,andsizeofthefaceareuncontrolled.Todothis,thenewmethodofdetectionproposedinthispaperisbasedprimarilyonatechniqueofautomaticlearningbyusingthedecisionofthreeneuralnetworks,atechniqueofenergycompactionbyusingthediscretecosinetransform,andatechniqueofsegmentationbythecolorofhumanskin.Awholeofpictures(facesandnofaces)aretransformedtovectorsofdatawhichwillbeusedforlearningtheneuralnetworkstoseparatebetweenthetwoclasses.Discretecosinetransformisusedtoreducethedimensionofthevectors,toeliminatetheredundanciesofinformation,andtostoreonlytheusefulinformationinaminimumnumberofcoefficientswhilethesegmentationisusedtoreducethespaceofresearchintheimage.Theexperimentalresultshaveshownthatthishybridizationofmethodswillgiveaverysignificantimprovementoftherateoftherecognition,qualityofdetection,andthetimeofexecution.
简介:Anewauditorysystemmodelbasedonacombinationofphysiologicalandpsycho-logicalacousticdatahasbeenproposed.Thismodelconsistsofabankofnonuniformbandpassfilters,detectorsandmain-frequencychoosingmechanisms,theyactasbasilarmembranes,innerhaircellsandnervefibers,respectively.Combiningwiththeimprovedcriticalbandwidthpa-rameters,theinputtothismodelisanalogoustothepressureattheeardrum,andtheoutputofthismodelsimulatesvariousfeaturesofthefiringpattern.Thesynthesizerobtainstheresul-tantspeechbyuseofthesimpleaddingmethod.Computersimulationsshowthattheresultantspeechishighlyintelligibleandnatural.Theproposedmodeliscorrectandtheimprovementofthecriticalbandwidthparametersiseffective.
简介:Thispaperisconcernedwithahighcharacteristicimageprocessingandrecognitionsystemthatisusedforinspectingreal-timeblemishes,streaksandcracksontheinnerwallsofhighaccuracypipes.Asaregulardetector,theBPneuralnetworkisusedforextractingfeaturesoftheimageinspectedandclassifyingtheseimages,ittakesfullyadvantageofthefunctionofartificialneuralnetwork,suchastheinformationdistributedmemory,largescaleself-adaptingparallelprocessing,highfault-tolerantabilityandsoforth.Besides,animprovedBPalgorithmisusedinthesystemfortrainingthenetwork,andmakingthelearningprocedureofthenetconvergestotheminimumofoverallsituationathighrate.
简介:Aneasyandpracticalsystemoffeeblefluctuationofliquidelectricalconductivityisdescribed.Thesystem,whichincreasesaprecisiononthethreeordersmagnitudeascomparedwiththatoftherecentdomesticconductivitymeter,isusedtomeasurethefeeblefluctuationoftheconductivitybymeansofcompensationandcomparison,sothatitsolves,theproblemofmeasuringthefeeblefluctuation.
简介:AtypeofdigitalchaoticeneryptionsystemwasproposedinRef.[1]whichusesaclassof1-Dpiecewiselinear(PWL)maptorealizechaoticencryptionanddecryptionsystemthroughtheinversesystemapproach.Inthegeneralstructureofencryptionsystem,adynamicalsystem∑(·)isusedtoconnectthelinearcombinationofn-orderdelaywiththeinputterminal.Inthispaperweshowthatthiscryptosystemcannotfrustratechosen-ciphertextattack.Atypeofchaoticencryptionsystembasedonself-synchronizingstreamcipherisproposed.Thissystemcanavoidchosen-ciphertextattackandhashighersecurity.