基于贝叶斯信息准则的极化干涉SAR图像非监督分类

An Unsupervised Classification Method of POLINSAR Image Based on Bayesian Information Criterion

该文提出一种利用贝叶斯信息准则自动确定聚类类别数的极化干涉SAR非监督分类算法。该方法首先利用Shannon熵特征对极化干涉SAR图像进行初始分类,然后利用期望最大化(Expectation-Maximization,EM)算法和标号代价(LabelCost)优化算法对分类结果进行迭代优化,同时通过贝叶斯信息准则(Bayesian InformationCriterion,BIC)自动确定非监督分类的最佳类别数。实验结果表明该算法能够较准确地确定分类类别数,并具有较为满意的分类效果。

An unsupervised classification algorithm established on the Bayesian Information Criterion （BIC） is presented for Polarimetric and Interferometric SAR （PolInSAR） images. First, an initial classification result is obtained by using Shannon entropy characteristic. Then, the result is optimized by Expectation-Maximization （EM） iteration algorithm and LabelCost optimization algorithm. Meanwhile, the method uses BIC to determine the number of clusters automatically. The experimental results show that the proposed method can not only obtain satisfied classification results, but also automatically determine the number of clusters.