This research proposes a deep learning-based resolution enhancement strategy for near-infrared II (NIR-II, 900?1880 nm) fluorescence imaging, which effectively overcomes the physical limitations of pixel quantity and size of InGaAs and HgCdTe detectors through fine-tuning the Real-ESRGAN super-resolution network model. The effectiveness of this method is validated in three typical scenarios: NIR-II whole-body vascular imaging in mice, multi-scene imaging in the NIR-IIc band, and NIR-II clinical application in diabetic foot assessment. The fine-tuned model significantly outperformed traditional bilinear and bicubic interpolation methods in both perception-based image quality evaluator (PIQE) and spatial resolution indicator—full width at half maximum (FWHM). Notably, the model demonstrated its capability to enhance the resolution of images with similar visual features, even when tested on untrained data from the NIR-IIc band, such as images of the mouse abdomens, intestines, and legs. This technology provides a new solution for high-quality NIR-II biomedical imaging, particularly offering clearer vascular visualization tools for clinical diagnosis, which is expected to play an important role in NIR-II fluorescence-guided clinical surgery.