The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLVIII-3-2024
https://doi.org/10.5194/isprs-archives-XLVIII-3-2024-273-2024
https://doi.org/10.5194/isprs-archives-XLVIII-3-2024-273-2024
07 Nov 2024
 | 07 Nov 2024

Identification of asbestos-cement roofing with the use of remote sensing data in the capital city of Poland

Małgorzata Krówczyńska and Ewa Wilk

Keywords: asbestos, asbestos recognition, remote sensing, Convolutional Neural Networks (CNN)

Abstract. Asbestos, a group of fibrous minerals prized for its strength, flexibility, and resistance to chemicals, heat, and electricity, was widely used in industrial applications, reaching its peak in the 1960s and 1970s with over 3,000 uses. Today, asbestos-cement roofs account for more than 80% of current asbestos-containing products. The International Agency for Research on Cancer has classified asbestos as a carcinogen, and the World Health Organization has called for global efforts to regulate and eliminate carcinogens, including asbestos, in both occupational and environmental exposure. Convolutional Neural Networks and high-resolution orthophotomaps were examined for the dense urban areas, i.e. the capital city of Poland, Warsaw. The area of Warsaw is divided into 18 districts. Two classes were distinguished: buildings with asbestos-cement roofs (“asbestos”) and buildings with other roof coverings (“non-asbestos”). Orthophotomaps, developed in 2021 based on aerial imagery with a spatial resolution of 5 cm and three spectral channels (RGB) were used. The overall accuracies (OA) over 90% were obtained. Differences in overall accuracy resulting from the size of the image signature were insignificant and amounted to up to 2 percentage points. The highest producer’s accuracies for the asbestos roofing class were obtained for a window of 128 by 128 pixels, while for the other roofing similar results were obtained for patterns with window sizes of 128 by 128 and 96 by 96 pixels. As a result of the research, the amount of asbestos-cement roofs was estimated, and their spatial location was determined in surveyed districts of the capital city of Warsaw. These results indicate that it is possible to train a neural network and, using available remote sensing data, estimate the scale of the phenomenon even in dense urban areas. Establishing a universally applicable technique would facilitate large-scale asbestos monitoring and management, thereby contributing to improved public health and safety across the EU.