Mamba-Based Residual Network for Remote Sensing Image Super-Resolution

Image super-resolution is the process of recovering high-resolution images from low-resolution inputs, which is crucial in low-level computer vision tasks. In recent years, progress in deep learning has significantly propelled the development of image super-resolution technology. Various models, such as convolutional neural networks (CNNs), generative adversarial networks, Transformers, and diffusion models, have been employed to address this problem with significant impact. Among these models, CNNs are effective in capturing local features but have limitations in modeling long-range dependencies; Transformers, despite their powerful modeling capabilities, face computational bottlenecks due to the quadratic complexity of the self-attention mechanism when dealing with large images. To address these issues, we propose the Residual Mamba Block, which combines state space models (SSMs) with residual connections, providing a linear complexity attention mechanism that effectively enhances long-range dependency modeling. Experimental results demonstrate the effectiveness and superior performance of the proposed method. Furthermore, our research indicates that vision state space models have great potential to become a foundational framework for future super-resolution tasks.