EDiffSR: An Efficient Diffusion Probabilistic Model for Remote Sensing Image Super-Resolution

Yi Xiao,Qiangqiang Yuan,Kui Jiang,Jiang He,Xianyu Jin,Liangpei Zhang

2023-10-30

Abstract:Recently, convolutional networks have achieved remarkable development in remote sensing image Super-Resoltuion (SR) by minimizing the regression objectives, e.g., MSE loss. However, despite achieving impressive performance, these methods often suffer from poor visual quality with over-smooth issues. Generative adversarial networks have the potential to infer intricate details, but they are easy to collapse, resulting in undesirable artifacts. To mitigate these issues, in this paper, we first introduce Diffusion Probabilistic Model (DPM) for efficient remote sensing image SR, dubbed EDiffSR. EDiffSR is easy to train and maintains the merits of DPM in generating perceptual-pleasant images. Specifically, different from previous works using heavy UNet for noise prediction, we develop an Efficient Activation Network (EANet) to achieve favorable noise prediction performance by simplified channel attention and simple gate operation, which dramatically reduces the computational budget. Moreover, to introduce more valuable prior knowledge into the proposed EDiffSR, a practical Conditional Prior Enhancement Module (CPEM) is developed to help extract an enriched condition. Unlike most DPM-based SR models that directly generate conditions by amplifying LR images, the proposed CPEM helps to retain more informative cues for accurate SR. Extensive experiments on four remote sensing datasets demonstrate that EDiffSR can restore visual-pleasant images on simulated and real-world remote sensing images, both quantitatively and qualitatively. The code of EDiffSR will be available at <a class="link-external link-https" href="https://github.com/XY-boy/EDiffSR" rel="external noopener nofollow">this https URL</a>

Image and Video Processing,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

### Problems Addressed by the Paper

The paper primarily addresses the following issues:

1. **Improving the quality of remote sensing image super-resolution (SR)**:
- Existing Convolutional Neural Network (CNN) methods, although achieving significant success in remote sensing image super-resolution tasks, often suffer from over-smoothing when generating high-quality images.
- Generative Adversarial Networks (GANs) can generate complex details but are prone to unstable training, leading to undesirable artifacts.

2. **Proposing an efficient and stable Diffusion Probabilistic Model (DPM) method**:
- By introducing more prior knowledge and designing an Efficient Activation Network (EANet) to reduce computational costs and improve generation quality.
- Developing a Conditional Prior Enhancement Module (CPEM) to extract more useful information from low-resolution images, thereby generating higher quality images.

3. **Addressing the shortcomings of existing DPM methods**:
- Most DPM-based methods rely on large networks for noise prediction, which is inefficient in large-scale remote sensing scenarios.
- The proposed method achieves a more lightweight design and improves performance in practical applications by simplifying the channel attention mechanism and using simple gating operations.

Through these improvements, the paper aims to develop an efficient and effective remote sensing image super-resolution model capable of restoring visually pleasing high-resolution images on both simulated and real-world remote sensing datasets.