Innovation in heritage conservation: Second island and EDIH Adria successfully tested AI for automatic drywall detection

Drywalls, stone monuments of skill and effort are an important part of the Croatian landscape. Their mapping, crucial for conservation and analysis, is traditionally a slow process that relies on fieldwork and manual interpretation of aerial footage. The second island vl. Filip Bubalo, a craft dedicated to researching and protecting vernacular heritage, recognized this challenge as an opportunity to apply new technologies. They asked a clear question: Can modern artificial intelligence detect these subtle structures on complex LiDAR images and thus improve the documentation process?

Solution through Test-Before-Invest (TBI) program

To test this idea, Second Island vl. Filip Bubalo cooperated with the EDIH Adria consortium through the “Test Before Invest” (TBI) programme. The aim of the project was to explore the technical feasibility and develop a prototype AI model for automatic drywall detection. The focus was on the application of modern computer vision techniques to solve a specific problem: Identifying irregular, often eroded stone structures on LiDAR low-contrast data with high noise levels.

As part of the TBI activities, experts from the Juraj Dobrila University of Pula, a partner in the EDIH Adria consortium, worked closely with the craft Second Island vl. Philip Bubalo. Key steps included:

• Creating a dedicated dataset: The user conducted a precise annotation of over 3000 examples of drywalls on more than 700 LiDAR images, obtained in cooperation with the Public Institution for the Management of the Marjan Forest Park, creating the first such data set in Croatia, specifically intended for AI training.
• Application of OBB technique: Due to the elongated shape of the walls, YOLOv11 architecture with support for oriented frames (Oriented Bounding Boxes – OBB) was used, allowing the model to describe the position and angle of the detected structures more accurately.
• Strategic model optimization: Through a series of experiments, it was found that merging all types of drywalls into one class and training on high-resolution images (1280px) gives the most stable results.

Results and Benefits for the Second Island

The TBI project has successfully demonstrated that, despite the challenges, the application of AI for drywall detection is technically feasible and promising:

• Basic performance level established: YOLOv11 medium OBB model achieved precision (precision) of 47.7% and find (recall) of 47.5%. These results prove that the model is able to learn the visual characteristics of drywalls and represent a solid, measurable basis for further development.
• Defined potential of “AI assistant”: The prototype has shown the ability to function as a tool to speed up manual work. Instead of searching for blank maps, researchers can use AI for the “first pass” which automatically marks almost half of all potential locations, which can significantly speed up the mapping process.
• Key challenges and solutions identified: The project confirmed that the biggest challenge is the very nature of LiDAR data. This insight guides future efforts towards acquiring higher density LiDAR data (minimum 300 points/m2, but preferably more) as the most significant step to further improve accuracy. In addition, the quality of input data would certainly increase significantly if AI models were applied directly to the cloud of dots instead of LiDAR images.

Results and Benefits for the Second Island

This TBI project provided the Second Island with valuable insights:

• Importance of specialised data: Success lies in creating your own, high-quality data sets tailored to the specific problem.
• AI as a tool for acceleration, not replacement: The greatest strength of AI in heritage conservation is in empowering professionals, allowing them to work faster and more efficiently.
• A clear path for further development: The project resulted in concrete recommendations, from the procurement of better quality data to the development of a GIS plugin that would turn this prototype into a practical tool.

Crafts Second island vl. Filip Bubalo plans to use the results of this project as a basis for further development and integration of this solution in cooperation with scientific institutions and considers application within public institutions dealing with the protection of cultural heritage.

The other island and vl. Filip Bubalo hereby expresses his great gratitude to the Juraj Dobrila University of Pula and EDIH Adria Consortium for the extremely valuable cooperation within the Test Before Invest (TBI) program.Through joint work, we have successfully tested the application of artificial intelligence for automatic drywall detection on complex LiDAR images, which can be said to be a precedent for projects dealing with the study of cultural heritage.The results confirm that modern technologies, such as LiDAR, AI and spatial data processing software, can play a key role in preserving cultural heritage – accelerating mapping processes and opening new doors to digital tools for landscape research.
We especially emphasize the expertise and commitment of the research team from the Juraj Dobrila University, whose support was crucial in the development of the first AI model of its kind in Croatia.
We believe that this project represents an important step towards a sustainable and innovative approach to the protection of our vernacular heritage, and we look forward to further development and cooperation on future initiatives.

Second island - craft for research, consulting and film production, Vl. Filip Bubalo

Conclusion

Cooperation between crafts Other island vl. Filip Bubalo and EDIH Adria through the TBI program is a successful example of how modern technology can be applied to address specific challenges in the preservation of cultural heritage. This project not only confirmed the technical feasibility of AI drywall detection, but also created the basis for the development of new digital tools that will make it easier for future generations to study and protect the landscape. EDIH Adria continues to support innovators such as the Second Island, empowering them to test and apply advanced technologies that bring real value.