RNFI, RNSI and ISPS: a complementary evaluation system for super-resolution of remote sensing imagery with downstream task relevance
Accurate evaluation of Super-Resolution (SR) methods is crucial for remote sensing applications, particularly for urban analysis where building and land cover segmentation depend on image quality. Conventional metrics, such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM), assess reconstruction fidelity but overlook low-resolution (LR) information, relative improvement, and task relevance. To address these limitations, we propose a resolution-aware evaluation framework comprising three complementary metrics: Reference-Normalized Fidelity Index (RNFI), Reference-Normalized Structure Index (RNSI), and Indicator of SR Potential Superiority (ISPS). These metrics jointly assess fidelity to high-resolution (HR) references, structural preservation, and improvement over LR inputs. Experiments on the Massachusetts, Wuhan University, Waterloo, and Potsdam datasets demonstrate that the proposed metrics provide complementary measures that are more closely associated with downstream indicators, including overall accuracy, Intersection-over-Union (IoU), precision, recall, and F1-score, than conventional PSNR, SSIM, LPIPS, or FID. This framework provides a more comprehensive and practically relevant basis for evaluating SR in remote sensing, linking reconstruction quality with practical downstream utility.
Authors
- Meiliu Wu (ORCID: https://orcid.org/0000-0002-5246-4603)
- Brian Barrett (ORCID: https://orcid.org/0000-0002-4380-4020)
- Zhimeng He (ORCID: https://orcid.org/0009-0004-3872-0035)
- Yuwei Cai
Institutions
- University of Glasgow (GB)
Publication Details
- Journal
- European Journal of Remote Sensing
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1080/22797254.2026.2767342
- Primary Topic
- Advanced Image Processing Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00