欢迎来到天天文库
浏览记录
ID:36851354
大小:3.63 MB
页数:58页
时间:2019-05-16
《基于边缘惩罚TMF的无监督SAR图像多类分割算法》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、摘要摘要合成孔径雷达(SyntheticApertureRader,SAR)图像分割是图像目标识别与解译技术的重要环节,一直是雷达信号处理领域的热点。但是SAR侧视成像和相干成像的特点,决定了SAR图像中包含大量相干斑乘性噪声,信噪比低,这使得传统的图像处理技术很难应用。本文针对SAR图像非高斯、非平稳的统计特性,提出了一种基于边缘惩罚三重马尔可夫场(TripletMarkovfields,TMF)的SAR图像多类分割新算法。该算法依据边缘惩罚准则将基于局部边缘强度信息的惩罚函数引入到TMF势能函数中,在对SAR图像的非平稳性很好地建模的同时解决了T
2、MF算法在分割时出现的边界定位不准确的问题,然后对新势能函数下的目标函数进行优化,导出多重区域迭代合并的贝叶斯最大后验模型(Bayesianmaximumposteriorimodel,MPM)分割公式。本文对测试图和实测SAR图像进行了仿真,仿真结果和分析表明:与经典的马尔可夫随机场(MarkovRandomField,MRF)模型和近年来的TMF分割算法相比,本文算法在抑制斑点噪声的同时,有效地提高了SAR图像的分割精度,尤其是在弱边缘处的分割定位更加准确。关键词:SAR图像多类分割三重马尔可夫随机场新势畿函数边缘惩罚多重区域合并Abstract
3、SyntheticApertureRader(SAR)imagesegmentationisanimportantstageforSAP,images’recognitionandunderstanding,andtheresearchfortheSARimagesegmentationalgorithmhasbeenahotspot.However,accordingtothecharacteristicsofSARside-viewimagingandcoherentimaging,theimagecontainsalargenumberofmu
4、ltiplicativespecklenoise,signaltonoiseratioislow.Thesepresentproblemsforstandardimageprocessingtechniques.Inthisdissertation,weproposeanewunsupervisedmulti—classsegmentationofSAP,imagesusingthetripletMarkovfields(TMF)modelsbasedonedgepenalty,thenewalgorithmfusethetraditionalene
5、rgyfunctionofTMFmodelwiththeprincipleofedgepenalty,whichcouldpreventsegmentfromsmoothingacrossboundarieswhilemodelingthenon-stationarySARimagebetter.Thenweoptimizetheobjectivefunctionthatstemsfromthenewenergyfunctiontoobtainaniterativemulti-regioncombinedBayesianmaximumposterio
6、rimodel(MPM)segmentationequationforthenewsegmentationalgorithm.SimulateddataandrealSARimagesareappliedtoevaluatetheperformanceoftheproposedalgorithminsegmentation.ExperimentalresultsandanalysisindicatethatcomparedwiththeclassicalMarkovrandomfield(MRF)andtherecentTMFsegmentation
7、algorithm,theproposedalgorithmeffectivelyimprovesthesegmentationaccuracyoftheSARimagewhilereducingtheinfluenceofmultiplicativespecklenoise,withtheweakedgelocationbeingmoreaccurateespecially.Keyword:SARimagemulti-classsegmentationTripletMarkovrandomfieldnewenergyfunctionedgepena
8、ltymulti-regionmerging创新性声明秉承学校严谨的学风和优良的科学道德,本人声明所呈交的论
此文档下载收益归作者所有