Fighting natural disasters with AI and remote sensing data

However, as prof. Chanussot points out: “Having data is one thing. Optimal exploitation of this data is another”. Indeed, the colossal amount of data these remote sensing sources needs to be processed and sorted through. It’s a colossal undertaking for humans to do without help. “It so happens that we have a powerful tool at our disposal: artificial intelligence”. The specific objective of the Chair is to develop advanced processing methods that will allow the rapid extraction of relevant information pertaining to disaster risk and/or response. “The kind of data analysis method we need hasn’t been developed yet. There are some induced pivotal and technical challenges to be tackled”. In particular, the processing methods in question need to be able to handle data that is heterogenous in nature, in addition to being massive. There are different kinds of data a satellite or an aircraft can collect via remote sensing: aerial photos, of course, but also multispectral and hyperspectral images, electromagnetic waves, sound waves, etc. “The algorithms need to be able to sort through all this information to highlight the useful part and deliver it in the shortest possible time. For instance, it needs to make the difference between natural changes like seasonal colors and vegetation, and abnormal changes on the ground. The challenge is to obtain a high-level representation of information in order to set aside the variations that disturb the analysis”.

Working hand in hand with the new Chinese Ministry of Emergency Management

To help him in this endeavour, prof. Chanussot is working with the Chinese Ministry of Emergency Management, the result of the merger, in 2018, of emergency management departments in various ministries. “This new ministry has access to a lot of data. They are actually involved in the ‘International Charter - Space and Major Disasters'. When a disaster occurs, they gather data from spatial agencies and people handle the information by hand. It’s a tedious work, and it requires solid expertise in photo interpretation”. The objective of the Chair is to provide these people with tools to that will make them faster and more efficient. The idea is not to replace them, “it is unrealistic” says. The Chair holder, but to provide them with computer programs in other words, to help with diagnostics and organizing relief efforts. These algorithms will be based on cutting edge learning techniques, such as manifold and graph-based representations, and data fusion.

Eventually, the Chair’s research program also proposes to look at another type of data that is also plenty and under-exploited: crowdsourcing. “The analysis of social media data, such as tweets, is an insightful source of very local information, with very high semantic value. As a result, unprecedented quantities of digital data are available, with heterogeneous scales and modalities. It poses a challenge comparable to remote sensing in that it necessitates careful and extensive sorting. There’s a lot of ‘bad quality’ data, in the sense that it is unverified, uncensored, unformatted and the means of expression, non-traditional”.

With the present Chair, prof. Jocelyn Chanussot embarks on a one-off scientific adventure

20 years of experience in the design of data processing and machine learning algorithms for the analysis of remote sensing data have accustomed him to theoretical rather than applied research. “The approach we adopted here is the opposite of what I’ve done over the years. We start from a concrete problem, and we try to find theoretical solutions, and not the other way around”. Such a course of action, fuelled by close collaboration with practitioners, ensures that advancements will be made in check with reality. “I find it very exhilarating to know that my work will have an impact in such a major societal context. Comprehensive and timely coverage is a potential game changer for disaster management. We don’t expect to solution all the challenges it poses by the end of the 5 years, but we will undoubtedly make significant progress in the field”.

*According to a report by the UNISDR (the United Nations Office for Disaster Risk Reduction) and the CRED (Centre for Research on the Epidemiology of Disasters).

Open Access Research