SUNet comprises three core modules: shallow feature extraction (via 3×3 convolutional layer), Unet feature extraction (integrating Swin Transformer blocks and patch merging layers), and reconstruction ...