基于相似度网络融合的极化SAR图像地物分类

从极化合成孔径雷达(synthetic aperture radar, SAR)图像中提取多种特征向量堆叠成一个高维特征向量用于地物分类,将导致部分特征向量的分类能力减弱或丧失。针对此问题,将每种特征向量看作为不同视角数据,提出了一种基于一致相似度网络融合的极化SAR图像非监督地物分类方法。首先,将极化SAR图像进行过分割,基于超像素提取5种特征向量以构建5个相似度矩阵;其次,采用一致相似度网络融合多视学习算法生成融合的相似度矩阵;然后,基于该矩阵进行谱聚类;最后,提出一种分类后处理策略修正错分像素。仿真和实测极化SAR图像地物分类结果表明,该方法性能优于其他5种经典方法。

Terrain classification of polarimetric SAR images based on consensus similarity network fusion

A variety of feature vectors are usually extracted from a polarimetric synthetic aperture radar (SAR) image and stacked directly into a high dimension feature vector to classify the different terrains in polarimetric SAR images, which results in the loss of some feature vectors’ discriminability. To address this problem, each feature vector is regarded as data from a different view of the image in this paper. Firstly, the polarimetric SAR image is over segmented to obtain a number of superpixels, and five similarity matrices are respectively constructed from five feature vectors extracted from polarimetric SAR images based on superpixels. Secondly, consensus similarity network fusion, which belongs to the multi view learning algorithms, is used to generate a fused similarity matrix. Thirdly, spectral clustering is performed on the fused similarity matrix. Finally, a novel classification post processing strategy is proposed to correct the misclassified pixels. Extensive experimental results conducted on a simulated and a real world polarimetric SAR images demonstrate the superiority of the proposed method, compared with five other classical methods.