Super Resolution SAR

SR-SAR addresses the improvement of MILSAR image interpretability through a software-based deep learning super-resolution pipeline for METEKSAN Savunma Sanayi. The system is built around the Swin2-MoSE transformer architecture and uses a dual-model strategy: one model optimized for edge preservation and one for speckle suppression. Outputs are combined via a weighted blending mechanism controlled by the parameter α, enabling adaptive balance. The system produces 2× and 4× super-resolved SAR outputs with patch-based preprocessing and a GUI for visualization and export, demonstrating meaningful enhancement without modifying radar hardware.