資源描述:
《基于DBN模型的遙感圖像分類.pdf》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在工程資料-天天文庫(kù)。
1、計(jì)算機(jī)研究與發(fā)展DOI:10.7544?issn1000-1239.2014.20140199JournalofComputerResearchandDevelopment51(9):1911-1918,2014基于DBN模型的遙感圖像分類呂啟1竇勇1牛新1徐佳慶1夏飛21(國(guó)防科學(xué)技術(shù)大學(xué)計(jì)算機(jī)學(xué)院并行與分布處理國(guó)防科技重點(diǎn)實(shí)驗(yàn)室長(zhǎng)沙410073)2(海軍工程大學(xué)電子工程學(xué)院武漢430033)(lvqi@nudt.edu.cn)RemoteSensingImageClassificationBasedonDBNModel1,DouYong1,NiuXin1,X
2、uJiaqing1,andXiaFei2LüQi1(NationalLaboratoryforParallelandDistributedProcessing,SchoolofComputer,NationalUniversityofDefenseTechnology,Changsha410073)2(ElectronicEngineeringCollege,NavalUniversityofEngineering,Wuhan430033)AbstractRemotesensingimageclassificationisoneofthekeytechnolog
3、iesingeographicinformationsystem(GIS),anditplaysanimportantroleinmodernurbanplanningandmanagement.Inthefieldofmachinelearning,deeplearningisspringingupinrecentyears.Bymimickingthehierarchicalstructureofhumanbrain,deeplearningcanextractfeaturesfromlowerleveltohigherlevelgradually,andd
4、istillthespatio-temporalregularizesofinputdata,thusimprovetheclassificationperformance.Deepbeliefnetwork(DBN)isawidelyinvestigatedanddeployeddeeplearningmodel.Itcombinestheadvantagesofunsupervisedandsupervisedlearning,andcanarchivegoodclassificationperformanceforhigh-dimensionaldata.
5、Inthispaper,aremotesensingimageclassificationmethodbasedonDBNmodelisproposed.Thisisoneofthefirstattemptstoapplydeeplearningapproachtourbandetailedclassification.Six-dayhigh-resolutionRADARSAT-2polarimetricsyntheticapertureradar(SAR)datawereusedforevaluation.Experimentalresultsshowtha
6、ttheproposedmethodcanoutperformSVM(supportvectormachine)andtraditionalneuralnetwork(NN).Keywordsremotesensingimage;syntheticapertureradar(SAR);landcoverclassification;deeplearning;restrictedBoltzmannmachine(RBM);deepbeliefnetwork(DBN)摘要遙感圖像分類是地理信息系統(tǒng)(geographicinformationsystem,GIS)的關(guān)
7、鍵技術(shù),對(duì)城市規(guī)劃與管理起到十分重要的作用.近年來(lái),深度學(xué)習(xí)成為機(jī)器學(xué)習(xí)領(lǐng)域的一個(gè)新興研究方向.深度學(xué)習(xí)采用模擬人腦多層結(jié)構(gòu)的方式,對(duì)數(shù)據(jù)從低層到高層漸進(jìn)地進(jìn)行特征提取,從而發(fā)掘數(shù)據(jù)在時(shí)間與空間上的規(guī)律,進(jìn)而提高分類的準(zhǔn)確性.深度信念網(wǎng)絡(luò)(deepbeliefnetwork,DBN)是一種得到廣泛研究與應(yīng)用的深度學(xué)習(xí)模型,它結(jié)合了無(wú)監(jiān)督學(xué)習(xí)和有監(jiān)督學(xué)習(xí)的優(yōu)點(diǎn),對(duì)高維數(shù)據(jù)具有較好的分類能力.提出一種基于DBN模型的遙感圖像分類方法,并利用RADARSAT-2衛(wèi)星6d的極化合成孔徑雷達(dá)(syntheticapertureradar,SAR)圖像進(jìn)行了驗(yàn)證.實(shí)驗(yàn)表明,
8、與支持向量機(jī)(SVM)及