简介:Arobustneuralnetworkcontroller(NNC)ispresentedfortrackingcontrolofunderwatervehicleswithuncertainties.ThecontrollerisobtainedbyusingbacksteppingtechniqueandLyapunovfunctiondesignincombinationwithneuralnetworkidentification.Modelingerrorsandenvironmentaldisturbancesareconsideredinthemathematicalmodel.Atwo-layerneuralnetworkisintroducedtocompensatethemodelingerrors,whileH∞controlstrategyisusedtoachievetheL2-gainperformance.Theuniformlyultimatelybounded(UUB)stabilitiesoftrackingerrorsandNNweightsareguaranteedthroughtheproposedcontroller.Anon-lineNNweightstuningalgorithmisalsoproposed.Goodperformancesofthetrackingcontrolsystemareillustratedbytheresultsofnumericalsimulations.
简介:Aparallelneuralnetwork-basedcontroller(PNNC)ispresentedforthemotioncontrolofunderwatervehiclesinthispaper.Itconsistsofareal-timepart,aself-learningpartandadesired-stateprogrammer,anditisdifferentfromnormaladaptiveneuralnetworkcontrollerinstructure.Owingtotheintroductionoftheself-learningpart,on-linelearningcanbeperformedwithoutsampledatainseveralsampleperiods,resultinginhighlearningspeedofthecontrollerandgoodcontrolperformance.Thedesired-stateprogrammerisutilizedtoobtainbetterlearningsamplesoftheneuralnetworktokeepthestabilityofthecontroller.Thedevelopedcontrollerisappliedtothe4-degreeoffreedomcontroloftheAUV'IUV-IV'andissuccessfulonthesimulationplatform.Thecontrolperformanceisalsocomparedwiththatofneuralnetworkcontrollerwithdifferentstructuressuchasnormaladaptiveneuralnetworkanddifferentlearningmethods.Currenteffectsandsurgevelocitycontrolarealsoincludedtodemonstratethecontroller'sperformance.ItisshownthatthePNNChasagreatpossibilitytosolvetheproblemsinthecontrolsystemdesignofunderwatervehicles.
简介:AnewanalyticalmethodusingBack-Propagation(BP)artificialneuralnetworkandkineticspectrophotometryforsimultaneousdeterminationofironandmagnesiumintapwater,theYellowRiverwaterandseawaterisestablished.Byconditionalexperiments,theoptimumanalyticalconditionsandparametersareobtained.Levenberg-Marquart(L-M)algorithmisusedforcalculationinBPneuralnetwork.Thetopologicalstructureofthree-layerBPANNnetworkarchitectureischosenas15-16-2(nodes).Theinitialvalueofgradientcoefficientμisfixedat0.001andtheincreasefactorandreductionfactorofμtakethedefaultvaluesofthesystem.ThedataareprocessedbycomputerswithourownprogramswritteninMATLAB7.0.Therelativestandarddeviationofthecalculatedresultsforironandmanganeseis2.30%and2.67%respectively.Theresultsofstandardadditionmethodshowthatforthetapwater,therecoveriesofironandmanganeseareintherangesof98.0%-104.3%and96.5%-104.5%,andtheRSDisintherangeof0.23%-0.98%;fortheYellowRiverwater(LijindistrictofShandongProvince),therecoveriesofironandmanganeseareintherangesof96.0%-101.0%and98.7%-104.2%,andtheRSDisintherangeof0.13%-2.52%;fortheseawaterinQingdaooffshore,therecoveriesofironandmanganeseareintherangesof95.3%-104.8%and95.3%-104.7%,andtheRSDisintherangeof0.14%-2.66%.Itisfoundthat21commoncationsandanionsdonotinterferewiththedeterminationofironandmanganeseundertheoptimumexperimentalconditions.Thismethodexhibitsgoodreproducibilityandhighaccuracyinthedeterminationofironandmanganeseandcanbeusedforthesimultaneousdeterminationofironandmanganeseintapwaterandnaturalwater.ByusingtheestablishedANN-catalyticspectrophotometricmethod,theironandmanganeseconcentrationsofthesurfaceseawaterat11sitesinQingdaooffshorearedeterminedandtheleveldistributionmapsofironandmanganesearedrawn.
简介:Thegeophysicalmodelfunction(GMF)describestherelationshipbetweenbackscatteringandseasurfacewind,sothatwindvectorscanberetrievedfrombackscatteringmeasurement.TheGMFplaysanimportantroleinoceanwindvectorretrievals,itsperformancewilldirectlyinfluencetheaccuracyoftheretrievedwindvector.Neuralnetwork(NN)approachisusedtodevelopaunifiedGMFforC-bandandKu-band(NN-GMF).EmpiricalGMFCMOD4andQSCAT-1areusedtogeneratethesimulatedtrainingdata-set,andGaussiannoiseatasignalnoiseratioof30dBisaddedtothedata-settosimulatethenoiseinthebackscatteringmeasurement.TheNN-GMFemploysradiofrequencyasanadditionalparameter,soitcanbeappliedforbothC-bandandKu-band.Analysesshowthattheσ0predictedbytheNN-GMFiscomparablewiththeσ0predictedbyCMOD4andQSCAT-1.AlsothewindvectorsretrievedfromtheNN-GMFandempiricalGMFCMOD4andQSCAT-1arecomparable,indicatingthattheNN-GMFisaseffectiveastheempiricalGMF,andhastheadvantagesoftheuniversalform.
简介:Oceanunderwaterexplorationisapartofoceanographythatinvestigatesthephysicalandbiologicalconditionsforscientificandcommercialpurposes.Andvideotechnologyplaysanimportantroleandisextensivelyappliedforunderwaterenvironmentobservation.Differentfromtheconventionalmethods,videotechnologyexplorestheunderwaterecosystemcontinuouslyandnon-invasively.However,duetothescatteringandattenuationoflighttransportinthewater,complexnoisedistributionandlowlightconditioncausechallengesforunderwatervideoapplicationsincludingobjectdetectionandrecognition.Inthispaper,weproposeanewdeepencoding-decodingconvolutionalarchitectureforunderwaterobjectrecognition.Itusesthedeepencoding-decodingnetworkforextractingthediscriminativefeaturesfromthenoisylow-lightunderwaterimages.Tocreatethedeconvolutionallayersforclassification,weapplythedeconvolutionkernelwithamatchedfeaturemap,insteadoffullconnection,tosolvetheproblemofdimensiondisasterandlowaccuracy.Moreover,weintroducedataaugmentationandtransferlearningtechnologiestosolvetheproblemofdatastarvation.Forexperiments,weinvestigatedthepublicdatasetswithourproposedmethodandthestate-of-the-artmethods.Theresultsshowthatourworkachievessignificantaccuracy.Thisworkprovidesnewunderwatertechnologiesappliedforoceanexploration.
简介:Owingtothelongpropagationdelayandhigherrorrateofacousticchannels,itisverychallengingtoprovidereliabledatatransferforunderwatersensornetworks.Moreover,networkcodingisprovedtobeaneffectivecodingtechniqueforthroughputandrobustnessofnetworks.Inthispaper,weproposeaReliableBraidedMultipathRoutingwithNetworkCodingforunderwatersensornetworks(RBMR-NC).Disjointmulti-pathalgorithmisusedtobuildindependentactualpaths,ascalledmainpaths.Somebraidedpathsoneachmainpatharebuiltaccordingtothebraidedmulti-pathalgorithm,whicharecalledlogicpaths.Whenadatapacketistransmittedbythesenodes,thenodescanemploynetworkcodingtoencodepacketscomingfromthesamegroupinordertofurtherreducerelativityamongthesepackets,andenhancetheprobabilityofsuccessfuldecodingatthesinknode.Braidedmulti-pathcanmakethemainpathstobemultiplexedtoreducetheprobabilityoflongpaths.ThispapermainlyemployssuccessfuldeliveryratetoevaluateRBMR-NCmodelwiththeoreticalanalysisandsimulationmethods.TheresultsindicatethattheproposedRBMR-NCprotocolisvaluabletoenhancenetworkreliabilityandtoreducesystemredundancy.