图像的超分辨率重建技术，是指对低分辨率图像进行上采样处理，并且在低分辨率图像的帮助下构建高分辨率图像的技术。此技术可以在模糊的像素信息中添加更多的细节，从而得到更加清晰的高分辨率结果图片。随着近几年来卷积神经网络技术的高速发展，使用卷积神经网络的单幅图像超分辨率重建方法也取得了令人满意的效果。使得图像重建的结果在细节上更加完善。

   由遥感卫星将摄像设备带入高空运行轨道，并进一步获取地面影像的过程即为遥感影像的获取过程。而遥感卫星所能直接获得的多数影像，受卫星本身硬件设备制约，很难直接得到工程期望的高清图像结果，不能直接用于实际需求。解决该问题主要有两种思路，一是直接提升卫星的硬件设备，二是采用超分辨率重建技术对遥感影像进行重建。而遥感卫星承重有限，而且一经发射难以收回，故而提升卫星本身硬件设施成本极高，故图像超分辨率重建技术对于遥感图像的实际应用意义重大。本文主要研究以深度学习为核心的单幅图像超分辨率重构技术在遥感影像超分辨率重建方面的应用，实现遥感影像的超分辨率重建，从而获得高清的遥感影像结果图片。

   本文采用残差网络模型进行遥感影像的超分辨率重建研究，并对重建后的结果与基于经典直连网络模型、SRCNN模型的重建结果进行对比，着重讨论重建结果的相似度与峰值信噪比，得到较为清晰的重建结果图片。此外，本文尝试将注意力机制模块与图像超分辨率重建技术相结合，在残差卷积模型中引入空间注意力区块，进一步优化图像的重建结果。

Image super-resolution reconstruction technology refers to the technology of upsampling low-resolution images and constructing high-resolution images with the help of low-resolution images. This technique adds more detail to the blurred pixel information, resulting in sharper, high-resolution images. With the rapid development of convolutional neural network technology in recent years, the single image super-resolution reconstruction method using convolutional neural network has also achieved satisfactory results. It makes the image reconstruction result more perfect in detail.

The process of taking the camera equipment into the high-altitude orbit by the remote sensing satellite and further acquiring the ground image is the process of acquiring the remote sensing image. However, most of the images that can be directly obtained by remote sensing satellites are restricted by the hardware equipment of the satellite itself, and it is difficult to directly obtain the high-definition image results expected by the project, and cannot be directly used for actual needs. There are two main ideas to solve this problem. One is to directly upgrade the hardware equipment of the satellite, and the other is to use super-resolution reconstruction technology to reconstruct remote sensing images. However, remote sensing satellites have a limited load-bearing capacity and are difficult to recover once launched. Therefore, the cost of upgrading the hardware facilities of the satellites is extremely high. Therefore, image super-resolution reconstruction technology is of great significance for the practical application of remote sensing images. This paper mainly studies the application of the single image super-resolution reconstruction technology with deep learning as the core in the super-resolution reconstruction of remote sensing images, and realizes the super-resolution reconstruction of remote sensing images, so as to obtain high-definition remote sensing image result pictures.

In this paper, the residual network model is used to study the super-resolution reconstruction of remote sensing images, and the reconstructed results are compared with the reconstruction results based on the classic direct connection network model. The similarity and peak signal-to-noise ratio of the reconstruction results are discussed, to get a clearer picture of the reconstruction result. In addition, this paper attempts to combine the attention mechanism module with the image super-resolution reconstruction technology, and introduce spatial attention blocks into the residual convolution model to further optimize the image reconstruction results.