Abstract: Aiming at the classification of polarimetric Synthetic Aperture Radar (SAR) images, a new approach using spectral clustering was propsed. The propsed method combined the polarimetic similarities of targets and spectral theory. It took two scattering parameters of the similarity characteristics of the target as input and used two-dimensional graph weights function instead of one-dimensional graph weights function to seek power value. It made the sampling points classification and the feature vector classification consistent to realize the classification of polarimetic SAR image. Experimental results show that the classification got by the method above is better than that got by K-means methods.

Key words: spectral classification, K-means algorithm, polarimetric similarity parameter, similarity, sampling

摘要： 针对极化SAR图像分类存在的问题,提出了基于SAR目标的极化特征的二维谱聚类方法。该方法可以充分考虑目标的极化相似性特征,利用二维的谱聚类方法实现极化SAR图像的分类。它以两目标散射的极化相似性参数图像作为输入特征,用二维图权函数代替一维图权函数求权值,使采样点分类和特征矢量分类相一致,从而实现极化SAR图像的分类。实验结果表明,该方法具有更好的分类结果,明显优于K均值分类。

关键词: 图谱聚类, K均值算法, 极化相似性参数, 相似性, 采样