基于精细去取向角的PolSAR图像非监督分类方法

Unsupervised Classification for Polarimetric Synthetic Aperture Radar Imagery Based on Refined Deorientation

针对基于模型分解的全极化合成孔径雷达非监督分类方法存在的大取向角的建筑物会被错误分类的问题,基于精细去取向角和复Wishart分类器,提出了一种全极化合成孔径雷达图像非监督分类方法,可以提高大取向角建筑物的正确分类率。首先,得到斑点滤波之后的相干矩阵数据,对目标相干矩阵进行精细去取向角;其次,基于去取向角之后的相干矩阵,实施Freeman和Durden提出的三分量模型分解,并作为初分类的结果;最后,基于模型分解结果,采用复Wishart分类器进行图像中目标分类,把目标分成15类。实验结果验证了该方法可以提高大取向角建筑物的正确分类率。

In the unsupervised classification of polarimetric synthetic aperture radar images, the largely oriented buildings may be wrongly classified as forests. Based on the refined deorientation and the complex Wishart classifier, an unsupervised classification method for polarimetric synthetic aperture radar images is proposed. First, the refined deorientation operation is implemented on the speckle filtered coherercy matrix. Second, the Freeman-Durden three-component model-based decomposition is carried out on the deoriented coherency matrix. Third, the complex Wishart classifier is implemented, and fifteen clusters are finally obtained. Experimental results demonstrate the increase of the percentage of the correct classification of largely oriented buildings.