Freqformer: Frequency-Aware Single Image Super Resolution Network for Biomedical HSI

Hyperspectral Imaging (HSI) provides rich spectral information that supports a wide range of biomedical applications, yet its spatial resolution remains inherently limited by sensor design and acquisition time. Single Image Super Resolution (SISR) offers a computational solution to enhance HSI spatial detail without compromising spectral integrity. Despite recent progress in HSI-SISR, existing models underutilize frequency-domain cues and often fail when applied to biomedical datasets. This paper presents FreqFormer-HSR, a frequency-enhanced reconstruction framework that augments spatial-spectral feature extraction through lightweight frequency modulation and multi-scale residual refinement. We evaluate the proposed method on two biomedical HSI datasets using a consistent experimental pipeline and compare against representative baselines. Quantitative and visual analyses demonstrate that FreqFormer-HSR achieves higher spatial fidelity while maintaining low spectral distortion. The results highlight the challenges of biomedical HSI-SISR across domains and show that frequency-aware feature modeling offers a promising direction for high-quality HSI reconstruction in clinical and preclinical imaging workflows. The full implementation and extended material are available at: github.com/NourO93/FreqFormer-HSR.