欢迎来到天天文库
浏览记录
ID:37391802
大小:8.19 MB
页数:127页
时间:2019-05-23
《基于形态学和神经网络的SAR影像专题信息提取方法研究》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、同济大学土木工程学院博士学位论文基于形态学和神经网络的SAR影像专题信息提取方法研究姓名:王栋申请学位级别:博士专业:摄影测量与遥感指导教师:陈映鹰20090501摘要离斑点降噪与目标分割处理效果。3.根据脉冲耦合神经网络(PCNN)神经元点火周期理论,构建了脉冲频率矩阵(PFM),提出了PCNN.PFM改进模型,克服了传统PCNN不能有效地处理具有强烈相干斑噪声的SAR影像的缺陷,有效改进了单极化SAR影像目标分割处理的效果和稳定性。4.研究集成学习理论框架,对影像分割得到的目标区域采用多学习机集成进行专题目标识别,有效提
2、高了系统识别的精度和泛化推广能力。5.集成以上改进算法模型,构建了基于形态学和神经网络的SAR影像专题信息提取原型系统,实现SAR影像专题目标的高效快速提取,并已在雷达景象匹配参考图制备和十一五村镇专题信息快速提取项目中得到实际应用。关键词:合成孔径雷达,目标提取,形态学,脉冲频率矩阵,脉冲耦合神经网络,盲信号处理,自组织映射神经网络,集成学习IIAbstractABSTRACTAstherapiddevelopingofSARimagingtechnology,agreatdealofresearchonSARisbein
3、gcarriedout.TheSARsystemhasdevelopedfromtheoriginalsingle-bandandsingle-polarizationSARtothemulti·bandandmulti-polarizationSAR。eventothefull-polarizationSAR.Theamountofdatastored,transmittedandhandledhasbeengrowingexponentiallywiththeincreaseofspatialandtimeresolut
4、ionoftheSARimage.ButtheSARimageprocessingisanewtechnology.ItismuchmoredifficulttoprocessforSARimagesthanforopticsimagesbecauseofthegreateffectbythecoherentspeckleandgeometricfeatures.ThetechnologyofSARimageprocessing,especiallyrealtimeandautointerpretationislaggedb
5、ehind.Thetraditionalmethodofvisualinterpretationandmanualprocessingcannotadapttothehugequantityofdata.Itbecomesabottleneckproblemtogetinformationfromthemagnanimousdata.HowtomakeUSeofthecomputertechnologytoimplementthequickunderstandingandinterpretationhasbecomeadif
6、ficultandkeyproblemurgenttoberesolved.Thispaperservesthepurposeofradarguidanceandsupportedbyaprogram”researchonthepreparationofreferencemapforradarscenematching'’consignedbytheXXXXacademeofspaceflightLtd.Tocreatingreferencemapformatchingthereal-timemap,thetypicalta
7、rgetinSARimageryshouldbeextractedfirst.Thisisalsoakeyproblemoftheprogram.Forautoandquickprocessing,thealgorithmshouldbeeffectiveandself-adaptive.Inthispaper’theauthorusingsomeself-adaptiveandparallel-processingtechnologysuch0.5themathematicalmorphology,self-organiz
8、ingmapneuralnetworks,pulse-coupledneuralnetworks(PCNN),blindsignalprocessingandensemblelearningtobuildupaframeworksofthematicinformationextractio
此文档下载收益归作者所有