资源描述:
《BP nerual-Effects of the number of hidden nodes used in a structured-based》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、NeuralComput&Applic(2009)18:249–260DOI10.1007/s00521-008-0177-3ORIGINALARTICLEEffectsofthenumberofhiddennodesusedinastructured-basedneuralnetworkonthereliabilityofimageclassificationWeibaoZouÆYanLiÆArthurTangReceived:13November2006/Accepted:8February2008/Publi
2、shedonline:27February2008ÓSpringer-VerlagLondonLimited2008AbstractAstructured-basedneuralnetwork(NN)withchoiceforthenumberofhiddennodesfortheimagebackpropagationthroughstructure(BPTS)algorithmisclassificationwhenastructured-basedNNwithBPTSconductedforimageclas
3、sificationinorganizingalargealgorithmisapplied.imagedatabase,whichisachallengingproblemunderinvestigation.ManyfactorscanaffecttheresultsofimageKeywordsHiddennodesclassification.OneofthemostimportantfactorsistheBackpropagationthroughstructureImageclassification
4、architectureofaNN,whichconsistsofinputlayer,hiddenNeuralnetworkFeaturessetlayerandoutputlayer.Inthisstudy,onlythenumbersofnodesinhiddenlayer(hiddennodes)ofaNNareconsid-ered.Otherfactorsarekeptunchanged.Twogroupsof1Introductionexperimentsincluding2,940images
5、ineachgroupareusedfortheanalysis.TheassessmentoftheeffectsforthefirstImagecontentrepresentationisachallengingproblemingroupiscarriedoutwithfeaturesdescribedbyimageorganizingalargeimagedatabase.Mostoftheapplicationsintensities,and,thesecondgroupusesfeaturesdesc
6、ribedrepresentimagesusinglow-levelvisualfeatures,suchasbywaveletcoefficients.Experimentalresultsdemonstratecolour,texture,shapeandspatiallayoutinaveryhighthattheeffectsofthenumbersofhiddennodesonthedimensionalfeaturespace,eithergloballyorlocally.reliabilityofc
7、lassificationaresignificantandnon-linear.However,themostpopulardistancemetrics,suchasWhenthenumberofhiddennodesis17,theclassificationEuclideandistance,cannotguaranteethatthecontentsarerateontrainingsetisupto95%,andarrivesat90%onthesimilareventhoughtheirvisualfea
8、turesareverycloseintestingset.Theresultsindicatethat17isanappropriatethehighdimensionalspace.Withastructured-basedneuralnetwork,theimageclassificationusingfeaturesdescribedbyindependentcom