Satellite imageries are an important geoinformation source for different applications in the Earth Science field. However, due to the limitation of the optic and sensor technologies and the high cost to update the sensors and equipments, the spectral and spatial resolution of the Earth Observation satellites may not meet the desired requirements. Thus, Remote Sensing Image Super-resolution (RSISR) which aims at restoring the high-resolution (HR) remote sensing images from the given low-resolution (LR) images has drawn considerable attention and witnessed the rapid development of the deep learning (DL) algorithms. In this research, we aim to comprehensively review the DL-based single image super-resolution (SISR) methods on optical remote sensing images. First, we introduce the DL techniques utilized in SISR. Second, we summarize the RSISR algorithms thoroughly, including the DL models, commonly used remote sensing datasets, loss functions, and performance evaluation metrics. Third, we present a new multi-sensor dataset that consists of Very High-Resolution satellite images from different satellites of various landscapes and evaluate the performance of some state-of-the-art super-resolution methods on this dataset. Finally, we envision the challenges and future research in the RSISR field.

卫星图像是地球科学领域不同应用的重要地理信息来源。然而，由于光学和传感器技术的限制以及更新传感器和设备的高成本，地球观测卫星的光谱和空间分辨率可能无法满足预期的要求。因此，旨在从给定的低分辨率（LR）图像中恢复高分辨率（HR）遥感图像的遥感图像超分辨率（RSISR）引起了广泛的关注，并见证了深度学习（DL）的快速发展。 ) 算法。在这项研究中，我们旨在全面回顾基于 DL 的单图像超分辨率 (SISR) 方法光学遥感图像。首先，我们介绍 SISR 中使用的 DL 技术。其次，我们彻底总结了 RSISR 算法，包括 DL 模型、常用的遥感数据集、损失函数和性能评估指标。第三，我们提出了一个新的多传感器数据集，该数据集由来自各种景观的不同卫星的超高分辨率卫星图像组成，并评估了一些最先进的超分辨率方法在该数据集上的性能。最后，我们设想了 RSISR 领域的挑战和未来研究。