Researcher in Residence programme explores uses of Artificial Intelligence (AI) in disaster management applications
The Satellite Applications Catapult has been applying Artificial intelligence (AI) and machine learning in a number of programmes, including collaborative work with the University of Oxford that has led to significant successes, such as identifying marine polluters through the Earth and Sea Observation System (EASOS).
AI and machine learning continue to hold huge potential for projects that use satellite data, particularly when the resulting information is required with low-latency and high-cadence. It’s a natural pairing on the surface, but in practice still requires a very high level of expertise to implement, which is why working with leaders in the field such as researchers from the University of Oxford is so valuable for both Catapult and our sector overall.
Enabling computers to learn how to extract information from satellite imagery
The EPSRC [1] funded Researchers in Residence (RIR) programme enables leading academics to be seconded into Catapults to undertake strategic project(s) and gain exposure to the business environment. As well as increasing knowledge exchange and co-creation between academia and the Catapult, they nurture the talent and skills development of researchers and Catapult staff, and act as a mechanism to rapidly accelerate research ideas and outcomes into potential new products and services.
Satellite Applications Catapult has been working with the University of Oxford for more than five years on AI and machine learning. Dr Steve Reece from Oxford’s Department of Engineering Science has worked with the Catapult on projects including the Earth and Sea Observation System (EASOS), funded by the UK Space Agency through the International Partnership Programme, and an Innovate UK-funded project to create an image labelling tool to aid machine learning. Now, as Researcher in Residence, Dr Reece is identifying opportunities to work with commercial partners on novel applications of machine learning and AI.
Satellites deliver huge volumes of data which, when combined with terrestrial information, can provide incredible levels of insight. However, obtaining information from satellite data requires intelligence, whether it is finding and classifying objects, identifying and learning patterns, or combining images to get the best result. Doing this takes specialist knowledge and time, but often the information is required as soon as possible so that it truly reflects the situation on the ground.
AI and machine learning provide a cost-effective solution to generating geospatial intelligence, including disaster management applications where they can analyse huge volumes of information produced immediately after an event by non-specialist volunteers.
For the EASOS Marine Watch application, Dr Reece developed a tool to model the dispersion of an oil slick backwards in time, as far 84 hours prior to its initial detection by satellite. Combined with vessel tracking data, this will enable government agencies to identify potential polluters, leading to increased fines and potentially reducing the number of polluting incidents by acting as a deterrent.
The RIR scheme continues to be a very successful model for nurturing Catapult-academia collaboration, accelerating the adoption of research by industry.
- Engineering and Physical Sciences Research Council