1222023RadioEngineeringVol.53No.1doi:10.3969/j.issn.1003-3106.2023.01.016引用格式:石雪,王玉.自适应空间约束融入混合模型的遥感图像分割[J].无线电工程,2023,53(1):122-128.[SHIXue,WANGYu.RemoteSensingImageSegmentationBasedonMixtureModelInfusedwithAdaptiveSpatialConstraint[J].RadioEngineering,2023,53(1):122-128.]自适应空间约束融入混合模型的遥感图像分割石雪,王玉(桂林理工大学测绘地理信息学院,广西桂林541004)摘要:为了降低图像噪声的影响并提高遥感图像分割精度,提出了一种自适应空间约束融入混合模型的遥感图像分割算法。考虑到学生t分布具有重尾特性比高斯分布更具有鲁棒性,利用学生t混合模型(Student's-tMixtureModel,SMM)建模像素光谱测度概率分布。为了避免图像噪声对分割结果的影响,基于马尔可夫随机场利用局部像素类属概率定义组份权重,将像素空间相关性融入SMM,进而构建出空间约束图像分割模型。为了实现自适应平滑系数的模型参数求解,采用梯度下降方法求解分割模型。采用本文算法对添加噪声的遥感图像进行分割实验,结果表明,所提算法可有效降低图像噪声的影响,同时可准确分割高分辨率遥感图像。关键词:图像分割;高分辨率遥感图像;学生t混合模型;马尔可夫随机场;自适应空间约束中图分类号:TP751文献标志码:A开放科学(资源服务)标识码(OSID):文章编号:1003-3106(2023)01-0122-07RemoteSensingImageSegmentationBasedonMixtureModelInfusedwithAdaptiveSpatialConstraintSHIXue,WANGYu(CollegeofGeomaticsandGeoinformation,GuilinUniversityofTechnology,Guilin541004,China)Abstract:Toreducetheinfluenceofimagenoiseandimprovethesegmentationaccuracyofimage,aremotesensingimagesegmentationalgorithmbasedonamixturemodelinfusedwithadaptivespatialconstrainisproposed.TheStudent's-tdistributionhasalongertailandismorerobustthanGaussiandistributionforimagenoise.ThedistributionofpixelspectralmeasurementismodeledbyStudent's-tMixtureModel(SMM).Inordertoavoidtheinfluenceofimagenoiseonsegmentationresults,thecomponentweightisdefinedbyMarkovrandomfieldusingtheattributeprobabilityoflocalpixels,andthespatialcorrelationofpixelsisintegratedintoSMM.Then,thespatiallyconstrainedimagesegmentationmodelisconstructed.In...