Polarimetric SAR Image Super-resolution Based on Coded Polarimetric Contexture Matrix and a Dual-branch Network

Polarimetric synthetic aperture radar (PolSAR) can acquire target’s richer electromagnetic scattering information compared to single polarization SAR. Meanwhile, the resolution of PolSAR image is relatively low due to the limitation of the hardware system. High-resolution PolSAR image can present more detailed information, which is beneficial for subsequent applications such as target classification and detection. In this vein, it is of great importance to improve the resolution of PolSAR image under existing hardware conditions. To address this problem, this work proposes a PolSAR image super-resolution method based on coded polarimetric contexture matrix and a dual-branch network. Firstly, a new representation of PolSAR data with reduced dimension is developed, which is denoted as the coded polarimetric contexture matrix. It is able to simultaneously characterize the polarimetric information related to the electromagnetic scattering properties of the targets and the contexture information related to the spatial texture properties. Secondly, combining the traditional and dilated convolution kernels, a dual-branch structure is designed to extract the polarimetric and contexture information separately. Finally, the extracted features are fused for better super-resolution. Experiments on PiSAR and Radarsat-2 data validate the effectiveness of the proposed method, which outperforms the comparison method both in terms of quantitative metrics and visualization.