Image-based automatic detection of efflorescence on limestone heritage constructions: Application to the Monastery of Batalha

A breakthrough in the automated detection of efflorescence on limestone heritage structures has been achieved, demonstrating that material-specific training enhances segmentation performance. The study revealed that convolutional neural networks (CNNs) can accurately delineate efflorescence with a best Binary IoU score of 92%, and even a specific model achieved 76% accuracy. This advancement was validated using imagery from the UNESCO-listed Monastery of Batalha in Portugal, showcasing its real-world applicability. The research highlights that training dataset composition significantly influences segmentation performance, providing practical guidance for quantitative documentation and monitoring of efflorescence in limestone masonry. Unlike previous qualitative assessments, this methodology offers a reproducible, data-driven approach to heritage conservation.

Bourgeois, I., Silva, D., Ferreira, V., & Rodrigues, H. (2026). Image-based automatic detection of efflorescence on limestone heritage constructions: Application to the Monastery of Batalha. Journal of Building Engineering, 129, 116924. https://doi.org/10.1016/j.jobe.2026.116924