Polarization Super-Resolution Remote Sensing Imaging Based on Compressed Sensing and Deep Learning

Remote sensing is widely applied in environmental monitoring, earth observation, and military reconnaissance, with increasing demands for higher spatial resolution, improved contrast, and multidimensional feature acquisition. Polarization imaging, particularly enabled by division-of-focal-plane (DoFP) polarization detectors, offers advantages in material differentiation and structural enhancement due to its ability to capture intensity, angle, and degree of polarization simultaneously in a compact, snapshot configuration. However, current research and applications of DoFP-based polarization imaging still encounter several key challenges: First, spatial resolution degradation and inter-channel aliasing arising from pixel-level downsampling and field-of-view (FoV) misalignment; Second, the absence of physical priors in most deep learning models, which limits their interpretability and robustness; Finally, maintaining image quality under sparse sampling conditions remains a significant technical challenge. To overcome these challenges, we propose PCSGAN, a dual-stage polarization super-resolution framework that couples compressed sensing theory with deep generative modeling, enhanced by polarization degradation simulation and sparse regularization. The oringal DoFP imaging resolution itself is half that of the detector's pixel scale. In contrast, the proposed method achieves 4× super-resolution reconstruction with only 6.25% sampling (equivalent to a single snapshot), effectively doubling the native spatial resolution of the detector while achieving high-fidelity polarization information recovery. We validate our approach on three public datasets and one self-collected dataset. Compared with six state-of-the-art methods across 20 types of remote sensing scenes, PCSGAN consistently achieves the best average results, improving PSNR by 2.3–11.7% and SSIM by 0.1%–12.8%. Effectiveness of PCSGAN is further demonstrated through UAV-based remote sensing experiments, showing strong potential for real-time, high-precision polarization imaging applications.