Semi-Supervised Classification of PolSAR Data with Multi-Scale Weighted Graph Convolutional Network

In this paper, a semi-supervised classification method for polarimetric synthetic aperture radar (PolSAR) images through multi-scale weighted graph convolutional network (MWGCN) is proposed, which utilizes the symmetric revised Wishart distance for weighted graph representation, graph integration module to incorporate the multi-scale information embedded in different hops, and superpixels to reduce the memory requirements and computation burden. The experiments clearly demonstrate that the proposed graph convolutional network (GCN) based method is suitable for PolSAR image interpretation in non-Euclidean domain and can achieve state-of-the-art classification results compared with recently proposed auto-encoder (AE) based and convolutional neural network (CNN) based methods.