极化SAR图像的聚类序列投影寻踪模型方法

Sequential projection pursuit clustering model for POL-SAR data unsupervised classification

针对极化SAR数据的分类问题,提出了序列投影寻踪模型方法进行极化SAR数据的无监督分类。该方法的特点是利用目标散射的极化相似性参数来表征目标特征;通过遗传算法逐步给出投影寻踪方法中的最佳投影,以获取高维数据的一维投影特征;进而采用EM算法估计混合模型的参数;最后由Bayes决策准则实现分类。该文对旧金山湾地区的极化SAR数据进行分类,得到了好的分类结果,实例计算结果分别与采用强度特征的无监督分类结果和直接利用散射熵-散射角分类的结果进行了比较,说明新方法对于极化SAR数据的分类具有明显的优越性。

On research of the classification of POLarimetric Synthetic Aperture Radar （POL-SAR） data, a Sequential Projection Pursuit Model （SPPM） for unsupervised segmentation of the POL-SAR image is proposed in this paper. The features of the high dimension data are extracted out via orthogonal projection and the classification is accomplished by the Bayes decision rule. Also the similarity parameters between two-scatter matrixes are calculated and expressed as the characters of a target and form new target characterized data. The SPPM utilize new target-characterized data to classify the target into various subclasses. Good-classified results have been obtained for the POL-SAR data classification. The classified results using the SPPM for the similarity parameters are better than those of using the SPPM for the intense information and using the scatter-entropy and scatter-angle plane. It shows our proposed method is a SAR data. good method in classification of the POL-SAR data.