Satellite Imagery Super Resolution Using Classical and Deep Learning Algorithms

The single image super resolution is very essential and needed in applications like image analysis, recognition, classification, better analysis and diagnosis of complex structured images. The applications in the different tasks of satellite imagery, medical image processing, CCTV image analysis, and video surveillance where a zoom is required, the super resolution becomes crucial in this case for a particular region of interest. In this paper, we have analyzed both classical and deep learning algorithms of satellite image enhancement using super resolution approach. The main focus of this paper lies in the comparison results of existing image enhancement algorithm such us bicubic interpolation, discrete wavelet transforms (DWT) based algorithms and deep learning based EDSR and WDSR architectures. These Deep learning (DL)- based improvement technique is presented to increase the resolution of the low-resolution satellite images. When compared to the existing classical methodologies, the DL-based algorithms significantly improve the PSNR while appropriately enhancing the satellite image resolution.