Enlighten-GAN for Super Resolution Reconstruction in Mid-Resolution Remote Sensing Images

Previously, generative adversarial networks (GAN) have been widely applied on super resolution reconstruction (SRR) methods, which turn low-resolution (LR) images into high-resolution (HR) ones. However, as these methods recover high frequency information with what they observed from the other images, tend to produce artifacts when processing unfamiliar images. Optical satellite remote sensing are of a far more complicated scene than natural Therefore, applying previous especially mid-resolution ones, leads unstable convergence and thus unpleasing artifacts. In this paper, we propose Enlighten-GAN for SRR tasks large-size optical Specifically, design enlighten blocks induce network converging reliable point, bring Self-Supervised Hierarchical Perceptual Loss attain performance improvement overpassing loss functions. Furthermore, limited by memory, large-scale need be cropped patches get through separately. To merge reconstructed whole, employ internal inconsistency cropping-and-clipping strategy, avoid seam line. Experiment results certify that outperforms state-of-the-art in terms gradient similarity metric (GSM) Sentinel-2