Researchers from the Universitat Oberta de Catalunya (UOC) in Spain have developed a new system that is capable of detecting asbestos — a highly toxic building material — not yet removed from the roofs of buildings.

According to its developers, the software applies artificial intelligence (AI), deep learning and computer vision approaches to aerial photographs, using RGB images.

Data pre-processing and inference process overview. The left panel shows the two main data collection steps, while the central panel illustrates how buildings are isolated and centered owing to cadastral data. The right panel exemplifies, first, the classification task, which delivers a number in the range of [0,1] expressing the likelihood of the presence of asbestos in the image. The classification task undergoes a Grad-CAM analysis, delivering an interpretable heatmap to understand which part of the image is most responsible for the classification score. Source: Remote Sensing (2024). DOI: 10.3390/rs16081342

"Unlike infrared or hyperspectral imaging methods, our decision to train AI with RGB images ensures the methodology is versatile and adaptable. In Europe and many other countries around the world this type of aerial imaging is freely available in very high resolutions," explained the researchers.

To train the deep learning system so that it could detect roofs containing asbestos, the developers used thousands of photographs held by the Cartographic and Geological Institute of Catalonia. In total, 2,244 images were used, with 1,168 of those images positive for asbestos and 1,076 of those images negative for asbestos.

The developers explained that the software can determine if asbestos is present in images by assessing different patterns, including the color, the texture and the structure of the roofs, and the area surrounding those buildings.

The team noted that the deep learning system had a success rate of over 80% for identifying the building material that was banned more than 20 years ago. This is due to the material being linked to 100,000 deaths a year around the world, primarily from lung cancer, as well as its link to other conditions including pleural tumors and pulmonary fibrosis, according to the World Health Organization. As such, the legal target for removing asbestos from public buildings is 2028 while the target for private buildings is 2032.

An article detailing the team’s findings, “Explainable Automatic Detection of Fiber–Cement Roofs in Aerial RGB Images,” appears in the journal Remote Sensing.