Multi-frame remote sensing super-resolution method based on dual-stream feature enhancement

Multi-frame remote sensing super-resolution (MFRSR) uses the displacement and detail differences between images from different frames to generate a clearer and more detailed image. Existing MFRSR methods are limited to the optical flow generated by single feature extraction and low-resolution input frames, resulting in low utilization of image information and low quality of generated optical flow. To this end, we propose a novel multi-frame remote sensing super-resolution method based on dual-stream feature enhancement (DFEMFRSR). The proposed method consists of two key components: (i) a Dual-stream Multi-head Attention Feature Extraction Module (DMAFEM), which employs a dual-branch architecture combined with multi-head attention to capture both global context and local texture details more effectively; and (ii) a High-Precision Optical Flow Estimation Module (HPOFEM), which leverages super-resolved input frames to enhance motion estimation accuracy. By jointly improving feature expressiveness and flow estimation quality, our method fully exploits the inter-frame redundancy and structural consistency. At the same time, the performance of our experiments is also better than other methods.