Blind Super-Resolution of Remote Sensing Images for Various Degradation Scenarios
Bili Lin, Junming Lao, Wujian Ye, Yijun Liu · 2025
The spatial resolution of satellite images is significantly degraded due to limitations in optical imaging hardware and atmospheric conditions, restricting the broader application of remote sensing technology. The emergence of deep learning has significantly improved the performance of super-resolution algorithms for remote sensing images. However, it still struggles to reconstruct severely degraded images with high quality, as it fails to remove elements like noise and blur that impair the clarity of remote sensing images. Therefore, this paper proposes an efficient blind super-resolution method for single optical remote sensing images capable of both denoising and deblurring. First, based on the characteristics of remote sensing images, we design a complex degradation model to generate a low-resolution dataset that closely mimics real-world degradation. Next, a dual-pyramid hybrid channel feature extraction module is introduced to construct the generator, enabling it to reduce noise while extracting effective high-frequency details from the image. Finally, the degradation dataset, the generator composed of feature extraction modules, and the discriminator are integrated with specific training strategies and optimizers to train a high-quality model. Experimental results show that our method not only achieves super-resolution but also significantly enhances image clarity under complex degradation, outperforming existing blind super-resolution methods.