Single Remote Sensing Image Super-Resolution via Convolutional Neural Network and Diffusion Model
Fanen Meng, Zhiguo Jiang, Fengying Xie, Sensen Wu, Zhenhong Du, Haopeng Zhang · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · 2026
Recent diffusion generative model super-resolution (SR) methods have made great progress in remote sensing image quality enhancement. However, the representation learning capability of diffusion models has not been fully explored, especially for photorealistic reconstruction using richer prior information. In this paper, we propose an end-to-end remote sensing image super-resolution method (convolutional neural network-guided diffusion model super-resolution method, CNN-DiffSR). This method combines the initial SR image generated by a convolutional neural network (CNN) to guide the diffusion model super-resolution reconstruction, generating visually clear and semantically accurate image samples that are closer to high-resolution images. Experiments on multiple benchmark datasets show that our method can simultaneously restore remote sensing images with more accurate texture details and high visual perception quality. In addition, more decent results and higher classification accuracy (+6.6%) are obtained in real-world remote sensing image super-resolution experiments and downstream task performance improvement tests, indicating that the proposed method provides a potential solution for improving the readability of remote sensing images.