Infrared Thermography with Dimensionality Reduction and Super-Resolution Preprocessing for Enhanced Thermal-Anomaly Visualization in Heritage Masonry

Infrared thermography was applied to the Southern Song stone stele Reconstruction of Yilao Hall in the Tianyi Ge Museum, China, to visualize surface and near-surface thermal anomalies potentially associated with weathering, crack-like features, and structural discontinuities. A motor-driven X–Y gantry equipped with an uncooled longwave infrared camera and two halogen floodlamps raster-scanned the 1.80 m × 0.90 m monument at 84 measurement nodes. At each node, 20 s of heating followed by 20 s of passive cooling was recorded at 50 Hz. Principal component analysis condensed the thermographic sequence into representative component images, and frequency-domain filtering enhanced local edge features. The resultant thermographic images were further processed using bicubic interpolation, super-resolution convolutional neural network, and super-resolution generative adversarial network (SRGAN). Bicubic interpolation achieved the highest peak signal-to-noise ratio and structural similarity index measure because it preserved the smoothed pseudo-reference, whereas SRGAN achieved the highest edge-contrast ratio and clearer visual delineation of inscription strokes and candidate crack-like thermal patterns. Because no independent high-resolution thermal ground truth is available for the historical object, the enhanced features are interpreted as candidate diagnostic cues rather than confirmed physical defects. The workflow can support non-contact preliminary screening of heritage stone masonry, while further validation on additional materials and independently verified defects is required.