【摘要】 马尔可夫随机场(Markov Random Field,MRF)广泛用于处理遥感图像的分类问题,然而MRF在构建极化合成孔径雷达(Synthetic Aperture Radar,SAR)图像模型时未考虑其非平稳特性且对初始分类较为敏感,为此本文提出了一种基于加权合成核与三重马尔可夫随机场(Triplet Markov Field,TMF)的极化SAR图像分类方法.该方法依据训练样本在特征空间上的距离,提出了加权合成核函数权重系数的自适应确定方法以提高初始分类的精度和普适性;为充分考虑极化SAR图像的非平稳统计特性,利用TMF对极化SAR图像进行统计建模以实现贝叶斯分类.实验结果表明,与基于MRF的极化SAR图像分类方法相比,本文所提方法可获得更高的分类精度和更平滑的同质区域分类结果,而且本文方法能更好地保持图像边缘信息.更多还原

【Abstract】 Markov random field( MRF) is widely applied to remote sensing images classification. However,the MRFbased classification method does not take the nonstationarity properties of images into account w hen it models polarimetric synthetic aperture radar( Pol SAR) images,and is sensitive to the initial classification. Therefore,this paper proposes a classification method of Pol SAR image based on the w eighted composite kernel and the triplet M arkov field( TM F). Based on the distances betw een the features of training samples,w e compute the kernel w eights of the w eighted composite kernel for improving the accuracy and popularity of the initial classification. Then,taking the nonstationarity properties of Pol SAR images into consideration,the TM F is introduced to model the statistics of real Pol SAR images to realize the Bayesian classification. Experiments indicate that the proposed method can obtain higher classification accuracy and smoother homogeneous areas than the M RF-based Pol SAR image classification method. M oreover,the proposed method can get more accurate edge location.更多还原