Answers are grounded in this paper and cite their sources.
Free optical Earth-observation missions such as Sentinel-2 provide frequent coverage of offshore wind farms, but their 10 m spatial resolution limits the analysis of floating platforms, which often occupy only a few pixels. This study presents a cluster-conditioned enhancement framework that maps Sentinel-2 RGB imagery to a nominal 0.75 m grid using a fixed WDSR super-resolution model and a genetic algorithm that optimizes interpolation, filtering, contrast, sharpening, and post-processing parameters. GEOSAT-2 imagery at 0.75 m resolution is used to define appearance-based clusters, validate the enhancement strategy under controlled downsampling, and provide structurally similar comparison images for real Sentinel-2 cases. The objective is based on feature recoverability and geometric matchability. The method was evaluated using 42 GEOSAT-2 images and 246 real Sentinel-2 scenes over the WindFloat Atlantic floating offshore wind farm in Portugal. Because the FLANN-SIFT correspondence score forms part of the Sentinel-2 genetic-algorithm fitness function, an independent post-optimization evaluation was performed using AKAZE features, brute-force Hamming matching, and RANSAC geometric verification. Across the 246 scenes, the GA outputs produced a mean of 1.254 RANSAC-verified AKAZE inliers per GEOSAT reference, compared with 0.633 for linear interpolation and 0.664 for passive WDSR; mean inlier ratios were 0.306, 0.186, and 0.194, respectively. Paired improvements were significant against both baselines (Holm-adjusted p < 10^-19). These results indicate that GA-guided, cluster-conditioned enhancement improves feature recoverability and cross-sensor matchability beyond the SIFT-FLANN criterion used during optimization.
No takes yet. Share an insight, caveat, or question.
Vieira et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: