Unsupervised Classification for Polarimetric SAR Images Based on Cross-View Tensor Product Graph Diffusion

Meilin Li,Huanxin Zou,Zhen Dong,Juan Wei,Xianxiang Qin

DOI: https://doi.org/10.1109/radar53847.2021.10028622

2021-01-01

Abstract:The scattering feature plays a crucial role in the polarimetric synthetic aperture radar (PolSAR) image classification. In most cases, the commonly used feature concatenation is adopted to construct the similarity matrix, thus leading to the discriminability loss of some single-view features. To solve this problem, this paper proposes an unsupervised PolSAR image terrain classification method based on the cross-view tensor product graph (CV-TPG) diffusion. Multi-view learning can effectively integrate the data sets from different views, moreover, the diffusion based on the CV-TPG is capable of mining the intrinsic affinity along the manifold structure of data. First, the PolSAR image is over-segmented into many superpixels. Second, five feature vectors are extracted from the PolSAR image via superpixels to form three high-dimensional feature vectors, resulting in three corresponding similarity matrices with the Gaussian kernel. Third, multiple CV-TPGs are obtained by the tensor product operation based on three similarity matrices, then CV-TPGs are linearly fused for applying the diffusion process to achieve a more discriminative similarity matrix. Finally, spectral clustering based on the diffused similarity matrix is adopted to perform terrain classification. Extensive experiments conducted on the Oberpfaffenhofen data set demonstrate that CV-TPG diffusion can effectively combine the characteristic information of multiple views and achieve higher classification accuracy, compared to five other competitive state-of-the-art methods.