To process large amounts of data due to the short revisit cycle characteristics of micro-satellite synthetic aperture radar (SAR) systems and to compensate for performance degradation due to weight reduction and miniaturization, the target identification must be improved through super-resolution (SR) for low-resolution (LR) satellite SAR images. In this paper, we generate SR satellite SAR images using the real-enhanced SR generative adversarial networks (Real-ESRGAN) for LR satellite SAR images and present the similarity of the peak signal-to-noise ratio (PSNR) and the structural similarity index measure (SSIM) for the segmentation with corresponding pairs based on key-points. SAR images are characterized by the location and intensity of major scattering points, unlike optical images. Therefore, it is necessary to understand the characteristics of the key-points detection method by considering the removal of clutter noise in the background and speckle noise around the target, outline, or boundary. Scale invariant feature transform (SIFT), speeded-up robust feature (SURF), binary robust invariant scalable key-points (BRISK), and oriented features from an accelerated segment test and rotated binary robust independent elementary features (ORB) are used as key-points detection methods, and the segmentation is generated for a local area with corresponding pairs based on key-points using the random sample consensus (RANSAC) method. As a result of the experiment, in the SR satellite SAR image, the PSNR in the segmentation with key-point- based corresponding pairs was high at the average 0.3850dB, and the SSIM was low at the average 0.0718, but when the background was included, the PSNR was maximum at 29.6937dB, and the SSIM was maximum at 0.5396.