DC-Mamba: A Degradation-Aware Cross-Modality Framework for Blind Super-Resolution of Thermal UAV Images

The low resolution of thermal imaging from unmanned aerial vehicles (UAVs) poses a substantial obstacle to the understanding and analysis of ground targets. Utilizing readily available high-resolution visible images presents a promising solution to improve the quality of thermal UAV images. However, current methods primarily focus on simple degradation conditions, neglecting the complexity of real-world degradation scenarios, such as blur and noise, which fail to meet the demands of practical applications. In this paper, we introduce a Degradation-aware Cross-modality Mamba (DC-Mamba) framework to super-resolve (SR) thermal UAV images by integrating degradation information with cross-modality cues. Our approach begins with a self-supervised learning framework that extracts degradation information directly from input images. This information guides the restoration process through the designed degradation-aware modules, which enhance model sensitivity to distorted regions. Additionally, we incorporate a vision-focused state-space module (SSM) to capture long-term spatial dependencies, thereby improving feature adaptability. To address modality disparities, we develop a cross-modality feature integration framework that leverages visible cues at three levels (interaction, refinement, and enhancement) to improve thermal image reconstruction quality. Extensive experiments demonstrate that the proposed method outperforms current state-of-the-art SR methods, providing more realistic details and superior performance across multiple evaluation metrics.