基于多尺度贝叶斯网络的SAR图像分割

SAR image segmentation based on multi-scale Bayesian network

提出了一种多尺度贝叶斯网络模型和相应推断算法,并将其应用于合成孔径雷达(synthetic aperture radar,SAR)图像分割。首先根据SAR图像的多尺度序列构建多尺度贝叶斯网络模型;然后设计了模型估计的置信传播(belief propagation,BP)算法,该算法包括同尺度结点之间的信息传播、细尺度到粗尺度的信息传播和粗尺度到细尺度的信息传播;最后计算出细尺度隐含结点的最大后验概率(maximum a posteriori probability,MAP),实现SAR图像的分割。实验结果表明,与单尺度贝叶斯网络模型方法和基于条件迭代模式的Markov随机场模型方法相比,基于多尺度贝叶斯网络的SAR图像分割方法具有较好的分割效果。

A multi-scale Bayesian network model and the associated inference algorithm,as well as a novel method of synthetic aperture radar （SAR）image segmentation based on the multi-scale Bayesian network are proposed.Firstly,the multi-scale Bayesian network model is constructed according to the multi-scale sequence of the SAR image.Then,the belief propagation （BP）algorithm,which consists of transmission of information among node in the same scale,from the fine scale to the coarse scale,and from the coarse scale to the fine scale, is presented to estimate the parameters of multi-scale Bayesian network model.Finally,the maximum a posteriori probabilities （MAP）of the finest scale hidden nodes are obtained to segment the SAR image.Experimental results show that the segmentation results based on the multi-scale Bayesian network model is better than those based on the single-scale Bayesian network or the Markov random field method using the iterated conditional mode algorithm.