Can AI be used to future-proof built cultural heritage?

By 2050, the EU's Energy Performance of Buildings Directive (EPBD) aims for all buildings in the EU – new and old – to be zero-emission buildings, minimising energy demand and eliminating carbon emissions. What does this mean for the buildings that tell our history? Will everything from the red farmer's cottage to the 19th century stone houses that dot our cities be fitted with insulated windows and solar panels?

"The Energy Performance of Buildings Directive has been around for a long time, but it has now been revised and the National Board of Housing, Building and Planning is working to implement it in Swedish legislation by 2026. "One change compared to the past is that buildings with official protection, i.e. protected as a building monument or in detailed plans, can no longer be completely exempted when it comes to setting minimum energy performance requirements, but the requirements can be adapted," says Ebba Gillbrand, researcher at the National Heritage Board, and continues:

"We at the Swedish National Heritage Board see an opportunity to work to ensure that the directive is interpreted in a way that takes into account the cultural values of buildings. We can contribute knowledge about how to improve energy efficiency in a careful way that does not jeopardise cultural values."

"The inclusion of heritage buildings in the Directive is a good thing," adds co-author Therese Sonehag. "Now heritage buildings can continue to be attractive to own and manage, rather than being seen as an obstacle to sustainable development."

AI as a tool to classify buildings

"Our project with RISE was based on the fact that there is no complete register of buildings of cultural and historical value in Sweden, and that cultural values are often mistakenly included in the energy declarations that suggest what energy efficiency measures can be done in a building. In other words, it is currently difficult to get an overview of which buildings are covered by the adapted energy requirements," says Camilla Altahr-Cederberg, researcher at the Swedish National Heritage Board.

RISE researchers Tim Johansson and Mikael Mangold came to the Swedish National Heritage Board with an idea that could potentially solve this challenge.

"For many years we have discussed the possibility of using 360-degree images of the street environment to extract information about buildings. The first thing we did in the project with the Swedish National Heritage Board was to determine camera positions and angles to make a good selection of the material. Then began the hard work of downloading the images and compiling them into a large database," says Tim Johansson, Technical Doctor in Construction Production at RISE.

A quarter of a million images

250,000 images were eventually created, a database of hundreds of gigabytes. The plan was to use AI to identify and map buildings of cultural heritage value in Sweden.

The hope was that the AI would capture buildings that were not included in other databases, such as Räkna Q (the county councils' register of buildings that have been granted protection and caution under the Planning and Building Act) or the Swedish National Heritage Board's Bebyggelseregister, where municipalities themselves register buildings of cultural value.

Initially, the researchers tested simply asking an AI and other language models to rate the buildings from 1 to 100 based on their heritage values.

"Then we saw that the AI was taking shortcuts. It was better to try to get different parameters in different ways. Based on a draft checklist for describing a building's cultural value from the Swedish National Heritage Board, we developed at least 20-30 different parameters that the AI had to consider before it could finally categorise the building's cultural value. We asked about the style of the building, the number of floors and so on," says Tim Johansson.

"We had to keep in mind that the AI is very eloquent and often just guesses. To check its reliability, we took a closer look at the buildings that were flagged as having a relatively high cultural value, but which were not included in any of the existing registers. After a workshop with the Swedish National Heritage Board, we concluded that assigning some form of cultural value to the buildings was the right way to go," says Mikael Mangold, senior researcher at RISE's Urban Development Unit.

The method could be used by the public sector

The team can now conclude that AI can be used as a tool to better classify and manage buildings of cultural value. However, the method needs further work and validation. The Swedish National Heritage Board and RISE believe that the idea of using AI to analyse a database of 360-degree images of the street environment could also benefit other public sector organisations.

"Carrying out this project has helped us in our dialogue with the Swedish National Board of Housing, Building and Planning on establishing methods and definitions within the Energy Performance Directive. We also see potential to use the study as a starting point when we develop, together with the Swedish National Board of Housing, Building and Planning and the Swedish Energy Agency, various forms of support for energy experts and building permit officers, for example. We look forward to continuing to work with RISE to develop the methodology," says Ebba Gillbrand.

Read the report Aspects of cultural values: from computer vision and street images