<![CDATA[Diagnosing early chromatic decay on tropical heritage façades is challenging because subtle discoloration is often masked by variable illumination and heterogeneous material properties. This study addresses that gap with two objectives: (1) to develop a radiometry-aware hybrid framework for chromatic decay detection, and (2) to validate its robustness across four heritage façades in Semarang, Indonesia. The methodology integrates 2D radiometrically normalized photogrammetric texture atlases, multi-space color and texture descriptors (HSV, CIELAB, GLCM, LBP), hierarchical spectral clustering, and Random Forest refinement with expert annotations. On 2,480 annotated tiles, the hybrid approach achieved aggregate micro-F1 ≈ 0.86 (per-site 0.84–0.87), surpassing cluster-only baselines (0.80) and RF-only models (0.82). Calibration with isotonic regression yielded Brier scores of 0.11–0.13 and Expected Calibration Error (ECE) ≈ 0.05–0.07. Statistical robustness was supported by site-stratified bootstrap and Wilcoxon tests. The resulting calibrated decay maps enable prioritized inspections, evidence-based conservation, and monitoring of tropical heritage assets.]]>
Copyrights © 2025
