摘要
VesselMonitoringSystem(VMS)providesanewopportunityforquantifiedfishingresearch.ManyapproacheshavebeenproposedtorecognizefishingactivitieswithVMStrajectoriesbasedonthetypesoffishingvessels.However,oneresearchproblemisstillcallingforsolutions,howtoidentifythefishingvesseltypebasedononlyVMStrajectories.ThisproblemisimportantbecauseitrequiresthefishingvesseltypeasapreliminarytorecognizefishingactivitiesfromVMStrajectories.Thispaperproposesfishingvesseltypeidentificationscheme(FVID)basedonlyonVMStrajectories.FVIDexploitsfeatureengineeringandmachinelearningschemesofXGBoostasitstwokeyblocksandclassifiesfishingvesselsintoninetypes.ThedatasetcontainsallthefishingvesseltrajectoriesintheEastChinaSeainMarch2017,including10031pre-registeredfishingvesselsand1350unregisteredvesselsofunknowntypes.Inordertoverifytypeidentificationaccuracy,wefirstconducta4-foldcross-validationonthetrajectoriesofregisteredfishingvessels.Theclassificationaccuracyis95.42%.WethenapplyFVIDtotheunregisteredfishingvesselstoidentifytheirtypes.Afterclassifyingtheunregisteredfishingvesseltypes,theirfishingactivitiesarefurtherrecognizedbasedupontheirtypes.Atlast,wecalculateandcomparethefishingdensitydistributionintheEastChinaSeabeforeandafterapplyingtheunregisteredfishingvessels,confirmingtheimportanceoftypeidentificationofunregisteredfishingvessels.
出版日期
2019年02月12日（中国期刊网平台首次上网日期，不代表论文的发表时间）
