Multitemporal Difference and Dynamic Optimization Framework for Multiscale Motion Satellite Video Super-Resolution

Motion alignment is a critical task in satellite video super-resolution. Most existing methods rely on optical flow or deformable convolution for motion alignment. However, these methods perform poorly in handling complex satellite video scenes and diverse moving objects. To address this challenge, we propose an efficient super-resolution method based on a multitemporal difference and dynamic optimization framework. Specifically, we introduce a multilevel temporal difference analysis mechanism to rapidly capture comprehensive motion information. Based on the extracted temporal difference data, we further develop a temporal differences-guided dynamic routing optimization module (T-DROM) to extract multiscale motion information. In addition, we introduce a multiattention enhancement and correction module to refine the long-term temporal difference features from T-DROM and reduce accumulated errors. These enhancements help recover spatial details and improve spatio-temporal consistency. We conducted detailed ablation studies to validate our contributions and compared the proposed method with state-of-the-art video super-resolution approaches. Experimental results show that our method improves video reconstruction quality while achieving an effective balance between performance and efficiency.