Automated Mapping and Characterization of Asbestos Cement Roofs with AI Techniques in Areas with Different Degrees of Anthropization

Although all forms of asbestos have been banned in the European Union since 2005, asbestos fibers are still present in millions of buildings and infrastructures, killing more than 70,000 people a year in Europe. In Italy, article no. 10 of Law 257/1992 describes how to draw up a map of the risk of exposure on our territory and how to draw up regional asbestos plans. This study presents an automatic method for the identification of asbestos cement-cement (CA) coverings in urban, agricultural and industrial areas, using the potential of deep learning algorithms, remote sensing data and GIS tools to implement a specific convolutional neural network (CNN) model. A U-net has been trained for the automatic detection of CA cover within simple high-resolution RGB orthophotos (0.20 metres). The research activity has developed a powerful tool for asbestos mapping over an entire regional territory, capable of identifying and analyzing different materials and their spatial distribution, developing an automatic process that is fundamental in the field of environmental monitoring for health, environmental, occupational safety and social security measures.