A Hybrid Geometric–Radiometric Embedding for Intelligent Diagnostics of Tropical Heritage Infrastructure

Routine inspection of ageing transport and heritage infrastructure in tropical climates is challenged by fine-scale geometric damage and non-geometric deterioration (e.g. moisture ingress) that often elude geometry-only point-cloud methods. We propose 3DMulti-FPFHI-SEMARANG, a novel 3D multimodal descriptor that fuses Fast Point Feature Histograms (FPFH) with a local Three-Dimensional intensity histogram tailored for laser-return and visible-spectrum variability, embedded within the PatchCore anomaly-detection framework. We validate the approach on a purpose-built synthetic dataset (including explicitly modelled crack faces and LiDAR radiometry), repeated field scans of a historical masonry span (Berok Bridge, Semarang), and a full-scale experimental tunnel segment (Lawang Sewu). Across these benchmarks, 3DMulti-FPFHI-SEMARANG achieves average precision = 0.95, recall = 0.90 and F1 = 0.91 on synthetic and field datasets, and yields notable detection improvements over standalone FPFH and a competitive CPMF baseline (relative F1 gains of 10–18%). Importantly, the method improves localization of narrow fissures and robustly flags intensity-based anomalies (e.g. moisture), while keeping data demands modest—supporting practical, repeatable monitoring of tropical heritage assets. Code and anonymized datasets will be released upon publication.