Artificial intelligence to detect asbestos

Twenty years after it was banned in construction, asbestos remains a major public health problem. Estimates suggest that in Catalonia alone, asbestos accounts for more than four million tonnes of fibre cement and between 6,000 and 30,000 tonnes of other compounds used in the material, which according to the WHO causes 107,000 deaths worldwide every year due to lung cancer, pleural cancer and asbestosis (pulmonary fibrosis). A research team at the Universitat Oberta de Catalunya (UOC) and the company DetectA has launched a project to address one of the key issues in the fight against asbestos: identifying rooftops made of the material. It aims to develop a technological solution based on artificial intelligence and computer vision to automatically detect which rooftops have asbestos, using aerial images in the public domain.

“There is currently no protocol to identify asbestos in the territory, or any systematic way to carry out the process. The only way is visual identification, which taking into account the large number of affected buildings that still exist, involves a very large financial and personal expense,” explained Javier Borge Holthoefer, the lead researcher in the Complex Systems group (CoSIN3), of the Internet Interdisciplinary Institute (IN3), who together with gata Lapedriza, the lead researcher in the Artificial Intelligence for Human Well-being (AIWELL) research group, affiliated to the eHealth Center and the Faculty of Computer Science, Multimedia and Telecommunications, is coordinating this transdisciplinary project for the UOC.

Determining the extent of the problem

This situation is particularly problematic given the different European, state and local regulations, according to which a census of buildings containing asbestos must be created and all asbestos must be removed in the coming years. “All public buildings must have removed all their asbestos by 2028. And the same applies to private buildings by 2032. This is in addition to the new Waste Law approved by the Spanish Congress in Madrid, which states that all municipalities must have registered the presence of asbestos in their municipal boundaries before May 2023. The first step to getting rid of asbestos is to have in-depth knowledge of where it is and the condition it is in, and this is where our project comes in. We can’t solve a problem if we don’t know its extent,” said Csar Snchez, who founded DetectA with Carles Scotto.

Training algorithms with aerial images

Faced with this challenge, the new project takes advantage of the two UOC groups’ expertise in image analysis, computer vision and machine learning – an artificial intelligence technique that consists of designing computer systems capable of learning from data and subsequently making predictions about new data that it has not seen before. In fact, these researchers have used this type of technology in other applications in the past, such as for identifying safe areas for pedestrians in urban environments.

In this case, the idea is to train an algorithm so that it can recognize which rooftops have asbestos by observing aerial and satellite images of Catalonia. “Asbestos was used to make tanks, tunnels, balconies, pipes and many other types of construction, but it’s thought that most of it is in rooftops,” said Javier Borge.

When carrying out this project, the researchers start with a database of images of rooftops with and without asbestos in areas in the Barcelona metropolitan area, images compiled and verified by the company DetectA. “The first step in a project like this one is to have an unarguable fact, which in this case is confirmed photos of rooftops to be able to train the algorithm so it knows which features to look for in new unclassified images. The more you train it, the better it gets,” said the IN3 researcher.

With the verified data and applying various techniques of computer vision – a discipline that extracts the information contained in an image – the algorithm learns and refines the classification of photos of rooftops and ceilings of different buildings. In this process, the research team will also apply advanced computational models of deep learning, known as deep neural networks, which, in the words of gata Lapedriza, are “layer-based models with millions of parameters, which use recent improvements in computational calculation capacity to learn how to automate tasks based on large data sets”.

Overcoming technological challenges

These cutting-edge technologies make it possible to overcome some of the challenges of this project and of the field of computer vision itself, and achieve robust results. “Asbestos is present in all kinds of buildings, so the algorithm needs large quantities of data to be able to understand all environments and contexts, ranging from the buildings in a big city like Barcelona, the typical buildings found in a village on the coast or in the Pyrenees, to the factories in an industrial estate or farms in a rural area,” pointed out Javier Borge.

Among the intrinsic challenges of computer vision projects, the AIWELL researcher highlighted the difficulties involved in identifying images of the same place produced under different conditions. “It’s very easy for a human being to understand that two images are of the same place, even if the light is different or one has been produced in the rain and the other on a sunny day. However, it’s very difficult for a machine to identify that two images are of the same place if there are changes in lighting or changes in the weather conditions. That’s why we have to do many experiments with a lot of data to be able to generalize the results,” he stated.

The researchers aim to define the working protocol and test this model using “images of municipalities that it has never seen to gauge its percentage of success and obtain a proof of concept of the technology by late summer,” said Lapedriza.

Free images in the public domain

The images that the project will use to train the algorithm will come from the Cartographic Institute of Catalonia database, a repository that is public and free for any user. This is one of its advantages compared to other similar projects, as it makes technology cheaper and broadens the scope of the project and its scalability. “There are initiatives that are similar to ours which depend on more sophisticated images, such as those which use multispectral cameras, which reflect the properties of the terrain. Those are expensive to obtain, so the geographical scope is usually limited to some neighbourhoods in large cities. If you want to solve a widespread problem like asbestos in Catalonia, we don’t think that this type of image is a viable solution,” explained Javier Borge.

“Our solution doesn’t need special flights to obtain specific images and it means we can create the map of fibre cement rooftops in the country without allocating new resources to it,” concluded Csar Snchez.

This project supports the Sustainable Development Goals (SDGs) 3 (Good Health and Well-Being), 9 (Industries, Innovation and Infrastructure) and 11 (Sustainable Cities and Communities).

UOC R&I

The UOC’s research and innovation (R&I) is helping overcome pressing challenges faced by global societies in the 21st century, by studying interactions between technology and human & social sciences with a specific focus on the network society, e-learning and e-health.

The UOC’s research is conducted by over 500 researchers and 51 research groups distributed between the university’s seven faculties, the E-learning Research programme, and two research centres: the Internet Interdisciplinary Institute (IN3) and the eHealth Center (eHC).

The University also cultivates online learning innovations at its eLearning Innovation Center (eLinC), as well as UOC community entrepreneurship and knowledge transfer via the Hubbik platform.

The United Nations’ 2030 Agenda for Sustainable Development and open knowledge serve as strategic pillars for the UOC’s teaching, research and innovation. More information: research.uoc.edu #UOC25years