Physically Interpretable Unsupervised Thermographic Clustering for Structural Alteration Diagnostics in Ancient Jade Artefacts

Ancient jade artefacts often exhibit multilayered weathering, mineralogical heterogeneity and complex subsurface alteration that cannot be fully assessed through visual inspection. This study proposes a physically interpretable unsupervised thermographic framework by integrating pulsed thermography (PT), long-pulsed thermography (LPT) and Self-Organizing Map (SOM) clustering for structural diagnostics in jade. To enhance depth-sensitive feature separation, a differential Thermographic Signal Reconstruction (TSR) encoding strategy was adopted: raw TSR signals from PT preserved emissivity-dependent surface information, whereas first-derivative TSR signals from LPT suppressed surface effects and highlighted deeper thermophysical contrasts. Operating directly on high-dimensional thermal sequences, SOM preserved the topological evolution of thermal responses and enabled clustering of surface mineralisation, layered weathering and internal heterogeneity without labelled data. Validation on a reference sample confirmed the method’s ability to distinguish shallow and deep features. When applied to a Shang-dynasty jade dagger, the framework mapped spatial variations in alteration depth and identified structurally vulnerable regions, and most importantly it revealed previously undocumented subsurface traces consistent with hafting or use-related contact that recover material evidence of ancient handling which cannot be detected through conventional examination. The proposed workflow offers a non-invasive, label-free and physically interpretable diagnostic pathway for mineral-based cultural heritage and provides a transferable approach for analysing semi-transparent or heterogeneously weathered artefacts.