Domain‐Aware Implicit Network for Arbitrary‐Scale Remote Sensing Image Super‐Resolution

Arbitrary‐scale image super‐resolution (SR) methods based on implicit neural representation (INR) have recently shown superior performance among deep learning‐based SR methods and they are preferable to other SR methods in the field of remote sensing image processing because of their ability of restoring high‐resolution (HR) images with arbitrary integer or non‐integer scales. However, these methods directly sample images of different scale factors for training and neglect the characteristic of the training distribution that there exists domain shift between samples of various scale factors. In this work, a Domain‐Aware Implicit Network (DAIN) is proposed to handle the arbitrary‐scale SR task from the perspective of domain adaptation where we minimize the domain discrepancy explicitly in order to learn domain‐invariant features. Besides, an enhanced local attention module is designed to further improve the performance of the model. Experiments have been conducted to demonstrate that the proposed method could achieve comparable performance and visual quality to other methods.