A Study on Transformer-Based Super-Resolution Network Design Techniques Utilizing Low Earth Orbit Multispectral Satellite Images

Low Earth orbit satellites play a vital role in Earth environment monitoring and industrial applications through multispectral data. However, the limited resolution poses challenges for applications that require fine details and for real-time processing. To address these issues, this study proposes a transformer-based super-resolution deep learning network that leverages the multi-channel characteristics of multispectral imagery and captures inter-channel interactions. The proposed network and procedure were validated in terms of data processing efficiency, reconstruction performance, and processing speed, demonstrating their effectiveness.