Machine learning in Domino-E use cases

What exactly is machine learning (ML) and  how does Domino-E make use of it in planning, resource allocation and human-computer interactions? In what ways does it benefit users? In this article we focus on answering these and other questions regarding machine learning in Earth Observation. 

Earth Observation (EO) produces a vast quantity of often unstructured data. For the average user to decipher, but also apply this information in their business or public service, this information must be accessible and understandable. In order for that to happen Domino-E employs machine learning in its federative system. Within that system ML is used to adapt the planning and allocation of resources and improve human-computer interaction. 

Data, based on data

Machine learning is a branch of AI, where, as the name suggests, systems learn from data alone rather than explicit programming. The more data you input into a machine learning system, the more accurate data analysis, predictions or models you will receive. This is because the algorithms applied get ‘trained’ or iteratively improve making associations between different data sets. A key advantage of this type of technology is that priorassumptions about the data (such as its quantity) and theoretical models become less crucial in delivering useful results. What becomes much more important though, are learning algorithms and sets of representative data on which they can train. 

Machine Learning and Domino-E 

For Earth Observation, machine learning is essential in reducing time and effort needed for human operators to analyse huge volumes of data produced by satellites. This essentially means that information, which was previously unattainable and incomprehensible for average users such as businesses, public services, researchers or policy-makers, becomes accessible and usable. Machine Learning in EO can have a huge impact in technological progress, benefiting all stakeholders, from private entities through businesses to academia. That is why Domino-E employs machine learning in its federative layer, which is a centralized unit of EO data capture.

More specifically, as presented on the figure above, the federative layer includes a virtual assistant (VAS). The virtual assistant acts as a one-stop-shop that allows users to access vast amounts of EO data in one place. This virtual assistant is aimed to incorporate machine learning to be able to translate ‘natural language’ requests into requests to external information sources such as satellites and vice versa. Additionally VAS ensures that the end user receives outputs in understandable and accessible language, while satellites and other information sources receive requests in technical terms that they can process.  According to Daniel Novak from Airbus Defence and Space, ‘The ability of VAS to interpret the responses in-putted by end users in ‘natural language’ is especially important because not all end users know all of the technical terms and parameters’. Therefore VAS combines machine learning with manually user-created knowledge, in the forms of user requests. This, ’improves the interaction between humans and AI systems’ and therefore has a long term benefit for AI systems. 

Machine Learning, Domino-E and you.

Analyses, observations and reviews made by machine learning are extremely valuable to an average user. According to Gauthier Picard from ONERA the French Aerospace Lab ‘Within Domino-E, services which ensure high reactivity to user requests, guarantee successful download of information, tailor solution and responses to individual requests and allow for a dynamic processing time, all use machine learning and AI technologies’. In short, ML is used in the areas of work, which are directed towards overcoming current technological, architectural and economical ‘roadblocks’ of existing EO services. More specifically machine learning is applied to decipher the complex system and network of satellites and ground stations, how they are connected, what their parameters are, what they behaviours are and at what time periods etc. This is done so that the overall system, or the pathway from user request to delivery of images, can be improved, but also in order to develop algorithms and mechanisms, which can improve satellite image retrieval in general. That way a multi-mission federative system, which is aimed at ensuring the most optimal outcomes, can not only provide users with fast and efficient results, but also provide a blueprint for future EO services. 

Within Domino-E machine learning allows the project to develop innovative key technologies to reduce time between user request and image delivery, which brings both short term and long term benefits. However the technology is constantly improving and, of course, learning. Let’s see how far we can go!

Published