Deep Learning-Based Approach for Heritage Structure Conservation

Strategic Insights -

Investigators at Czech Technical University Prague have made significant contributions to photogrammetry remote sensing and spatial information sciences by developing an automated deep learning-based approach to segment stone and mortar in heritage masonry. This innovative method uses a U-Net convolutional neural network to accurately distinguish between stone and mortar in high-resolution imagery, showcasing its potential for scalable and objective digital heritage conservation. The researchers demonstrated the effectiveness of their approach on the iconic Old Town Bridge Tower in Prague, achieving a class-averaged F1 score of up to 85.58%.

Key Takeaways:

  • The study proposed a deep learning-based approach using a U-Net convolutional neural network to automatically segment stone and mortar in heritage masonry.
  • The researchers trained the U-Net model on high-resolution imagery of Prague's Old Town Bridge Tower, achieving a class-averaged F1 score of up to 85.58%.
  • The proposed method significantly improves segmentation speed and consistency over manual methods, enabling precise manual segmentation of all stones, excluding non-masonry features.
  • The created segmentation maps can be easily converted to finished vector drawings, supporting conservation tasks such as structural monitoring and damage detection.
  • The trained model will aid future documentation of the Charles Bridge and has potential applications in scalable, objective digital heritage conservation.
  • J. Vynikal, L. Beloch, and T. Boucek collaborated on the research, which was conducted at the Department of Geomatics, Faculty of Civil Engineering, Czech Technical University Prague.

Statistics:

  • Class-averaged F1 score: up to 85.58%
  • Ground sampling distance: 1 mm
  • Number of images used for training: not specified
  • Number of images used for testing: not specified
  • Dataset used: a collection of high-resolution images of the Old Town Bridge Tower and its interior
  • Scanner used: an RTC360 laser scanner
  • Drone used: a DJI M300 drone equipped with a P1 camera
  • Number of stones segmented: all stones in the dataset, excluding non-masonry features

Sources:

  • Automated Segmentation of Stone and Mortar in Heritage Structures: A Case Study on the Old Town Bridge Tower in Prague. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2025,XLVIII-M-9-2025():1587-1592.
  • The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - http://www.isprs.org/publications/archives.aspx
  • NewsRx. Czech Technical University Prague Researchers Highlight Recent Research in Photogrammetry Remote Sensing and Spatial Information Sciences (Automated Segmentation of Stone and Mortar in Heritage Structures: A Case Study on the Old Town Bridge ...). Science Letter. October 24, 2025; p 70.