Hyperspectral image upscaling works well on unseen sensors fast
Super-Resolving Unseen Hyperspectral Sensors at Any Scale via Spatial Operators
Computer Vision and Pattern Recognition
Summary
Hyperspectral images capture many colors beyond normal cameras but are often low resolution. Improving their resolution usually requires specific models trained for each camera and scale. The authors created a single model that learns general patterns to enhance images from many sensors and scales without retraining. Their method predicts spatial operators instead of direct color values, allowing smoother and faster image enhancement. Tests show it works better and faster on new cameras than previous approaches.
What this means in practice
- •For remote sensing teams: Improve spatial resolution of hyperspectral images from new sensors without retraining or extra data collection.
- •For environmental monitoring groups: Generate high-resolution hyperspectral images at multiple zoom levels efficiently for better analysis of land and vegetation.
Authors
Ji-Xuan He, Guohang Zhuang, Bo Junge, Tingyi Li, Lingchen, Miaomiao Cai, Yanan Qiao, Xiujin Liu, Junfeng Fang
Abstract
Achieving cross-sensor generalization and arbitrary-scale reconstruction with a single model remains challenging in hyperspectral super-resolution (HSR). Although recent methods support arbitrary-scale reconstruction, applying them to new sensors or scales beyond the training range often requires additional data and computation to maintain reconstruction quality. To address these challenges, we propose OmniHSR, which predicts band-shared spatial operators rather than spectral values. Cross-Spectral Mapping (CSM) resamples inputs with any number of bands to fixed reference positions and predicts local operators with Gaussian supports. Continuous Operator-Field Reconstruction (COFR) composes these operators into a continuous field and applies them to all original bands for arbitrary-scale reconstruction. Experiments demonstrate that operator prediction outperforms direct spectral-value prediction on all seven datasets. Trained solely on ARAD with only 0.538M parameters, OmniHSR outperforms all directly transferred baselines on six unseen datasets without target-domain training data or adaptation. Across twelve upsampling factors from $\times2$ to $\times48$, it improves average PSNR on Pavia U and Chikusei by 0.55 dB over the strongest baseline. It also surpasses baselines trained from scratch or adapted on the target sensor and achieves up to $36\times$ faster inference. Our code will be publicly released soon.