How much wider will a crack in a sandstone tower get in 50 years? This seemingly unanswerable question is being tackled by a group of French scientists using data and algorithms. The goal is not only preservation, but also transforming this specific data into concrete information that can influence policymakers and foster a sense of responsibility among the public.
Teaching machines to "see" instead of human eyes.
The real challenge isn't "using AI to photograph heritage," but rather how a machine can understand degradation, a concept that is inherently dependent on human perception, language, and perspective.
Ann Bourgès, a senior conservation scientist at the French Museum Research and Restoration Center of the French Ministry of Culture, laid the groundwork for the project. Since 2022, Bourgès and two colleagues have launched two doctoral projects with research students Adèle Cormier and David Roqui. The two pilot sites were deliberately chosen: the octagonal sandstone base of the Strasbourg Cathedral tower – a 13th-century Rayonnant Gothic structure that withstands harsh continental winters and scorching summers; and the Bibracte archaeological site near Autun in Burgundy – a Gaulish settlement first excavated in the late 19th century.
Roqui's mission was to teach AI not only to read data, but also to "see." According to The Art Newspaper , this meant training the model to identify cracks in photographs, then comparing two photos taken at different times to determine how much the crack had widened. The research team faced two major challenges: the ratio between global phenomena and the specific microclimatic characteristics of each heritage site, and the lack of standardization among commercial measurement devices. To overcome this barrier, the project used thermal infrared imaging – a technology that can reveal water seepage and mineral salt accumulation within rocks that are undetectable to the naked eye.
Initial results are very encouraging. According to the Peer Community Journal , the multimodal model tested on data from Strasbourg Cathedral achieved 76.9% accuracy and an F1 score of 77.0% – a 43% improvement over conventional AI architectures like VisualBERT or Transformer, and a 25% improvement over a pure PerceiverIO model. Even more noteworthy, when run individually, sensor data only achieved 61.5% accuracy while image data only reached 46.2% – demonstrating that the true power lies in combining both sources of information.
Global ambitions
The impressive technical figures are just the beginning. What Bourgès and her colleagues are aiming for is a much bigger ambition: to create a tool that any conservationist or archaeologist in the world can access, regardless of local or national budget.
According to The Art Newspaper , the entire methodology of the project will be published as open source and integrated into the Espadon platform – a national project initiated by the French Ministry of Culture to digitize heritage with augmented reality technology, while providing researchers with access to all known data on any building.
The ultimate goal, as clearly stated by Ms. Bourgès, is: "We want users to be able to visualize how their specific location will change over time in relation to the local climate." Instead of dense, data-driven scientific reports, the tool will create a visual representation: how much of this wall's plaster or paint will be lost after 100 years.
This is the dimension beyond pure science that Ms. Bourgès – also the Secretary-General of the French branch of the International Council on Monuments and Sites (ICOMOS) – emphasizes: "It is a means of gathering and clearly showing what the climate crisis is causing. If you can show people a picture of their wall losing half its plaster in 100 years, they will understand immediately." And according to her, that is also why the need for this type of tool is so great and urgent: "Whether you are a conservationist or an archaeologist, everyone wants to know what to do. But to know what to do, you need to know what is going to happen."
AI for heritage preservation: A pan-European picture
The French project is just one of many similar projects.
HYPERION, funded by the EU with nearly €6 million, is being piloted in Rhodes (Greece), Venice (Italy), Tønsberg (Norway), and Granada (Spain). HYPERION's unique feature is its integration of the community into the monitoring process via a mobile application, turning each passerby into a "living sensor." The YADES project, funded through the Marie Skłodowska-Curie Programme, focuses on heritage in Cyprus, Greece, and Italy, with an emphasis on 80 rotational trips between organizations, ensuring the technology remains integrated with the local community.
Three projects, three approaches - but the same understanding: AI cannot replace humans in cherishing heritage, but it can help humans better understand what is being lost, so that timely interventions can be made.
Source: https://baodanang.vn/khi-ai-giup-bao-ton-di-san-3339586.html
