"Space-FreqDiff SR"

"This dataset is derived from high-resolution remote sensing imagery acquired by the Gaofen-2 (GF-2) satellite, which provides sub-meter panchromatic data and multi-spectral observations with high spatial and temporal resolution. Representative scenes covering selected regions and time periods were collected and systematically preprocessed, including radiometric calibration, atmospheric correction, geometric correction, and image co-registration, to ensure radiometric consistency and geometric accuracy.To support super-resolution and image reconstruction research, a PAN-RS\u2013based degradation strategy was applied to the original high-resolution (HR) imagery. Specifically, the HR images were degraded through spatial blurring, down-sampling, and sensor-related response simulation to generate corresponding low-resolution (LR) images, while preserving realistic remote sensing characteristics. As a result, each LR image is strictly paired with its HR counterpart, forming well-aligned LR\u2013HR image pairs suitable for supervised learning tasks.The resulting dataset features high spatial fidelity and consistent LR\u2013HR correspondence, making it well suited for applications such as remote sensing image super-resolution, image fusion, target detection, and fine-scale land surface analysis. By releasing this dataset, we aim to provide a reliable benchmark for high-resolution remote sensing research and to facilitate data sharing and reproducible experimentation within the remote sensing and computer vision communities."

本数据集基于高分二号（Gaofen-2, GF-2）卫星获取的高分辨率遥感影像构建，该卫星可提供亚米级全色数据与多光谱观测数据，兼具优异的空间与时间分辨率。研究团队采集了覆盖选定区域与时段的典型场景，并开展系统性预处理工作，涵盖辐射定标、大气校正、几何校正与图像配准，以保障影像的辐射一致性与几何精度。为支撑超分辨率与图像重建相关研究，团队针对原始高分辨率（High Resolution, HR）影像采用了基于PAN-RS的降质策略。具体而言，通过空间模糊、下采样与传感器响应模拟对高分辨率影像进行降质处理，生成对应的低分辨率（Low Resolution, LR）影像，同时保留真实的遥感成像特性。由此，每一张低分辨率影像均与对应的高分辨率影像严格配对，形成对齐良好的LR-HR影像对，可适配监督学习任务需求。本数据集具备高空间保真度与稳定的LR-HR对应关系，可广泛应用于遥感影像超分辨率、图像融合、目标检测与精细尺度地表分析等场景。本数据集的发布旨在为高分辨率遥感研究提供可靠的基准数据集，并推动遥感与计算机视觉领域内的数据共享与可复现实验开展。