超分辨率技术可提升遥感图像空间分辨率，为基于遥感图像的目标检测、场景分类等任务提供更加清晰的数据集，具有广泛的应用价值。然而，现有基于深度学习的超分辨率方法存在监督次数不足的问题，导致超分辨率重建图像中易出现细节损失和伪细节。针对这一问题，本文提出基于分级监督的遥感图像超分辨率方法（MSSR）。首先，提出了一个分级监督网络架构，通过引入多级真值图像作为监督，为超分辨率过程提供充足的图像细节恢复指引，进而减少超分辨率结果中细节损失和伪细节的出现。其次，为了便于构建级数可变、超分辨率倍数可变的分级监督网络，设计了一个轻量化的、超分辨率倍数可灵活调整的同构超分辨率模块（BSRC）。各级BSRC的网络结构基本相同，便于迁移网络权重，缩短训练时间。最后，针对分级网络超分辨率倍数一定时，网络级数及各级分辨率倍数有多种组合方式的问题，对比多种分级方式下的超分辨率结果，给出最佳网络分级方式。此外，构建了一个包含世界各地复杂细节地面场景的遥感图像数据集（RSSRD）。在该数据集和UCMerced、AID两个公开数据集上进行超分辨率实验，实验结果显示本文方法优于现有常用超分辨率方法。

Objective: Super-resolution improves the spatial resolution of remote sensing images, providing more detailed information for various satellite applications. However, existing methods often generate pseudo-detail and lose true detail in reconstructed images due to insufficient supervision images. To address this issue, a progressive Super-Resolution method based on Multi-level Supervision structure was proposed (MSSR). Method: First, a multi-level supervision network structure was proposed to introduce more ground truth images as guidance, which reduced the loss of true detail and mitigated the appearance of pseudo-detail in the output image. The MSSR network consists of several Basic Super-Resolution Components (BSRCs) and multi-level supervision. The overall super-resolution scale factor of the MSSR network can be set flexibly. The BSRCs can be increased or decreased like building blocks, with fewer BSRCs for a smaller overall super-resolution scale factor and more BSRCs for a larger overall super-resolution scale factor. The scale factor of each BSRC is determined by the overall scale factor and the number of BSRCs. Second, a scale-factor-adjustable and lightweight basic super-resolution component was designed to enable the construction of multi-level supervision networks with different number of BSRCs and different scale factors like building blocks. The BSRC consists of a multi-scale feature extraction module, a global feature extraction module and an image reconstruction module. Considering that the scale factor of each BSRC should have some flexibility, the network structure of each BSRC is the same except for the image reconstruction module. This helps to shorten the overall training time of the network. Finally, a method of dividing super-resolution overall scale factor was proposed to explore the effects of different numbers of BSRCs and different scale factors of BSRCs on the performance of multi-level supervision network. For the super-resolution process with a certain scale factor, we need a method to divide the overall scale factor into each BSRC. As the number of BSRC increases, the number of supervision increases, and the super-resolution ill-posedness is reduced. At the same time, the total number of network layers increases, the number of computations increases. We determined the optimal number of BSRCs and their respective super-resolution scale factors by comparing the super-resolution effects of multiple BSRC combinations. Additionally, a new remote sensing dataset containing worldwide scenes was also constructed for the super-resolution task in this paper. Result: In order to adequately train and test the proposed super-resolution method and existing methods, we used our datasets and two existing super-resolution datasets: the UCMerced dataset and the AID dataset. We compared our method with the state of art methods: VDSR, SRGAN, RDN, RCAN, DRN and TransENet. The experiment results on three datasets demonstrated that our MSSR network outperformed the state-of-the-art (SOTA) methods. Conclusion: The progressive network and multilevel supervision structure can effectively suppress super-resolution ill-posedness and reduce pseudo-detail and detail loss in super-resolution results. By further analyzing the experimental results, we found that multi-level supervision has greater performance gains in super-resolution tasks with larger scale factor. We speculate that the multi-level supervision network will have better performance in super-resolution tasks above 4 times. In future research, we will experiment with multi-level supervision networks in super-resolution tasks at larger magnifications such as 5times, 8 times and so on.