Enhancing Remote Sensing Image Super-Resolution Guided by Bicubic-Downsampled Low-Resolution Image

Image super-resolution (SR) is a significant technique in image processing as it enhances the spatial resolution of images, enabling various downstream applications. Based on recent achievements SR studies computer vision, deep-learning-based methods have been widely investigated for remote sensing images. In this study, we proposed two-stage approach called bicubic-downsampled low-resolution (LR) image-guided generative adversarial network (BLG-GAN) super-resolution. The BLG-GAN method divides procedure into two stages: LR transfer and stage, real-world images are restored to less blurry noisy bicubic-like using guidance from synthetic obtained through bicubic downsampling. Subsequently, generated used inputs network, which learns mapping between corresponding high-resolution (HR) image. By approaching problem finding optimal solutions subproblems, achieves superior results compared state-of-the-art models, even with smaller overall capacity network. As utilizes bridge HR shows improved quality models trained learn direct an Experimental satellite datasets demonstrate effectiveness improving perceptual preserving fidelity.