A Summary of Super-Resolution for Satellite Videos Via Learning-Based Methods

With the development of remote sensing techniques, remote sensing data can be obtained with higher spatial, higher spectral, and higher temporal resolution. In addition, to get higher spatial resolution, super-resolution for increasing spatial resolution is getting special attention. In this paper, we will focus on some classical learning-based superresolution methods to investigate the adaptability for satellite videos with low imaging quality. Methods include sparse representation, collaborative representation, and deep learning methods. Experiments show that learning-based methods can perform well for single-frame super-resolution for satellite videos. Methods based on deep learning show higher PSNR and SSIM. And multi-frame super-resolution will be good for moving objects. However, it may also bring negative influence for a stationary scene, which is caused by low satellite video quality, such as winkling noise, a vibration of a camera, overexposure of metals.