A Super Resolution Method for Remote Sensing Images Based on Cascaded Conditional Wasserstein GANs

Abstract

High-resolution (HR) remote sensing imagery is quite beneficial for subsequent interpretation. Obtaining HR images can be achieved by upgrading the imaging device. Yet, the cost to perform this task is very huge. Thus, it is necessary to obtain HR images from low-resolution (LR) ones. In the literature, the super-resolution image reconstruction methods based on deep learning have unparalleled advantages in comparison to traditional reconstruction methods. This work is inspired by these current mainstream methods and proposes a novel cascaded conditional Wasserstein generative adversarial network (CCWGAN) architecture with the residual dense block to generate high quality remote sensing images. We validate the proposed method on the NWPU VHR-10 dataset. Experimental results show our CCWGAN method has superior performance compared with the state-of-the-art GAN methods.