欢迎来到天天文库
浏览记录
ID:34509329
大小:133.85 KB
页数:7页
时间:2019-03-07
《robust classification of variable length sonar sequences》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库。
1、RobustclassicationofvariablelengthsonarsequencesJoydeepGhosh,NarsimhamV.GangishettiandSrinivasaV.ChakravarthyDepartmentofElectricalandComputerEngineeringTheUniversityofTexas,Austin,TX78712-1084Abstract.Twotypesofarticialneuralnetworksareintroducedf
2、ortherobustclassicationofspatio-temporalsequences.TherstnetworkistheAdaptiveSpatio-TemporalRecognizer(ASTER),whichadaptivelyestimatesthecondencethata(variablelength)signalofaknownclassispresentbycontinuouslymonitoringasequenceoffeaturevectors.Ifth
3、econdenceforanyclassexceedsathresholdvalueatsomemoment,thesignalisconsideredtobedetectedandclassied.ThenonlinearbehaviorofASTERprovidesmorerobustperformancethantherelateddynamictimewarpingalgorithm.ASTERiscomparedwithamorecommonapproachwhereinaself
4、-organizingfeaturemapisrstusedtomapasequenceofextractedfeaturevectorsontoalowerdimensionaltrajectory,whichisthenidentiedusingavariantofthefeedforwardtimedelayneuralnetwork.Theperformanceofthesetwonetworksiscomparedusingarticialsonogramsaswellasfea
5、turevectorsstringsobtainedfromshort-durationoceanicsignals.1IntroductionManyintelligenttasksinvolvedecision-makingandoutputbehaviorinresponsetospatio-temporalstimuli.Therecognitionandnonlineartransformationofcontinuous-timesignalsorsignalsequencesist
6、husfundamentaltoawiderangeofcognitiveprocesses.Classicationofspatio-temporalsignalsisalsobasictomanyengineeringapplicationslikespeechrecognition,seismiceventdetection,sonarclassicationandreal-timecontrol[11,13].Thecentralissueintheprocessingofsuchs
7、ignalsishowpastinputsorhistory"isrepresentedorstored,andhowthishistoryaectstheresponsetothecurrentinputs.Temporalinformationcanberepresentedexplicitlybycreatingaspatial(static)representationoftemporaldata.Tradingtimewithspaceisthebasisfortimedelaye
8、mbeddingstructuresandfrequencyencoding.Alternatively,thepastcanbeimplicitlyrepresentedbyamemoryobtained,forexample,byinsertingtime-delaysintocellsorconnections,usingfeedbackorrecurrentconnections,oraddinginternalstatestotheprocessingcells[13].Thischa
此文档下载收益归作者所有