预览加载中,请您耐心等待几秒...
1/3
2/3
3/3

在线预览结束,喜欢就下载吧,查找使用更方便

如果您无法下载资料,请参考说明:

1、部分资料下载需要金币,请确保您的账户上有足够的金币

2、已购买过的文档,再次下载不重复扣费

3、资料包下载后请先用软件解压,在使用对应软件打开

基于卷积神经网络和迁移学习的结构损伤识别 Abstract Withthedevelopmentofintelligentdetectiontechnology,theapplicationofstructuraldamagerecognitionhasbeenfurtherexpanded.However,traditionalstructuraldamagerecognitionmethodshavesomeshortcomings,suchasthedifficultyinfeatureextractionandmodeloptimization.Inthispaper,weproposeastructuraldamagerecognitionmethodbasedonconvolutionalneuralnetwork(CNN)andtransferlearning.Firstly,wepreprocessthesensordatacollectedfromthestructureandextractthefeaturesofdifferentsensorclusters.Secondly,weusethepre-trainedCNNmodelfortransferlearningtoobtainastructuraldamagerecognitionmodel.Finally,weevaluatetheperformanceoftheproposedmethodusingexperimentaldata.Theexperimentalresultsshowthattheproposedmethodhasgoodaccuracyandrobustness,andhasbetterperformancethantraditionalmethods. Introduction Structuraldamagecanhaveseriousconsequences,andearlydetectionofstructuraldamageiscriticaltopreventingaccidents.Traditionaldamagedetectionmethodsmainlyrelyonmanuallyselectedfeaturesandmathematicalmodels,whichisacomplexandtime-consumingprocessandrequiresprofessionalexpertise.Tosolvetheseproblems,theapplicationofintelligentdetectiontechnology,representedbymachinelearning,hasbeengraduallyexpanded. Machinelearninghasbeenwidelyusedinstructuraldamagerecognition.Amongthem,convolutionalneuralnetworkisadeeplearningmethodthathasbeenusedinimagerecognition,speechrecognition,naturallanguageprocessingandotherfields.CNNischaracterizedbyitsabilitytoextracthigh-levelfeaturesfromrawdataandautomaticallyoptimizemodelparameters.TheessenceofCNNistouseconvolutionallayerstoextractlocalfeaturesofinputdata,andthenusepoolinglayerstocombinetheextractedfeaturesofdifferentregionstoformincreasinglyabstractfeatures,andthenusefullyconnectedlayersforclassificationandidentification.Transferlearningisamachinelearningmethodthatusesthepre-trainedmodeltolearnanewtask,whichcaneffectivelyreducetheamountoftrainingdataandimprovetheperformanceofthemodel. Inthispaper,weproposeastructuraldamagerecognitionmethodbasedonCNNandt