Self-adaptive Super-resolution for Remote Sensing Images

#remote #sensing #super-resolution

About the speaker: Hanlin Wu received the B.S. degree in statistics from Beijing Normal University, Beijing, China, in 2015, where he is pursuing the Ph.D. degree in computer application technology. His primary research interests include remote sensing image processing, super-resolution reconstruction, salient object detection, and deep learning technology.

Abstract: In remote sensing images (RSIs), the visual characteristics of different regions are versatile, which poses a considerable challenge to single image super-resolution (SISR). Most existing SISR methods for RSIs ignore the diverse reconstruction needs of different regions and thus face a serious contradiction between high perception quality and less spatial distortion. To address this problem, we propose a new saliency-guided feedback GAN (SG-FBGAN) to apply different reconstruction principles for areas with varying levels of saliency. Besides, most existing super-resolution (SR) methods only focus on specific integer scale factors (e.g., x2, x3, x4). Retraining a model for each scale factor is extremely time-consuming and takes up too much storage. Also, the scale factor should not be restricted to an integer but any positive real number. To this end, we devise a scale-aware dynamic network (SADN) for continuous-scale SR. The experimental results show that our proposed continuous-scale upsampling module can replace the previous fixed-scale upsampling layers and obtain a continuous-scale SR network while maintaining performance. Our SADN uses much fewer parameters and outperforms the state-of-the-art SR methods.