- Starion Italia initiates development of novel approach to artificial intelligence (AI) data processing to improve efficiency and performance.
- Funded by the Italian Space Agency (ASI), the two-phase project will apply a data-agnostic methodology to create an image processing chain that can be used multiple times in different contexts and domains.
Starion Italia has started development of a generic AI-based image processing software programme that will combine multiple steps in one reusable chain. The development of the Versatile Intelligent Processing Enhancement and Refinement chain (VIPER) project is in two phases and will focus on use cases in the space domain; however, the end result will be an adaptable AI application encompassing commonly repeated tasks that can be used with any computer vision systems, significantly speeding up the time required to develop AI solutions.
Initially, Starion’s AI experts will use VIPER to improve the efficiency and performance of existing systems within two use cases, demonstrating the capabilities and potential of this AI chain approach. Both involve the analysis and processing of space-related data: the first will use satellite Earth observation (EO) data to identify illegal landfills, while the second will focus in another direction, using telescope data to identify near-Earth objects (NEOs).
VIPER is being funded through ASI’s R&D projects and services for ’Robotic and artificial intelligence enabling technologies‘.
EO is a prime example of a sector where there are significant benefits in using AI-powered applications to process optical and radar data mainly, especially given the exponential growth of EO data in the last decade – one estimate is that approximately 100 petabytes of EO data are now generated every year[1]. Europe’s Copernicus programme alone produces over 20 terabytes of EO data daily. A 2024 briefing paper developed by the World Economic Forum (WEF) in collaboration with Deloitte identified three notable impacts of using AI with EO data: the ability of AI tools to answer complex questions more quickly and accurately; making EO insights available to non-expert users; and driving business model innovation. The WEF paper also highlighted potential positive links to climate-positive action[2].
Andrea Cavallini, Starion Competence Area Lead for ‘AI and Machine Learning’ and ‘EO and Downstream’, explains the impetus for developing VIPER: “Typically, every AI programme is developed from scratch and any tools that do exist only address one task. However, this fails to take into account the fact that some of the tasks involved in working with large datasets are common, such as reducing noise and enhancing resolution, and typically more than one is required each time. With VIPER, we will provide a versatile, reusable solution that efficiently combines these preprocessing tasks and can be implemented by users in a straightforward way, either using the complete chain or a subset of the modules provided.”
Reducing the cost and time to develop and implement AI programmes is increasingly desirable. Even for relatively modest AI development, estimates range from tens of thousands to over 1 million euros per project, depending on their features and complexity, with timeframes from 6 months to several years. What is unavoidable is the computational resources required, which scale based on the complexity of the algorithms and data involved, and the significant effort to prepare and label the datasets, which often requires a manual process. There are new tools offering automated annotations, but human intervention remains crucial.
Marco Di Clemente, ASI Head of Technology Development and Space Design, notes: “What makes VIPER especially valuable to us in this context is its potential to avoid ‘reinventing the wheel’. Many AI systems tend to reuse similar tools and architectures, but often train on different data. VIPER will streamline this by providing a more standardised approach, saving both time and resources in development, which is why we see great potential in this project.”
The first phase for VIPER has been allocated 18 months and will involve development of the processing chain and training for the two use cases as proof of concept on ground. In the 12-month second phase, the software will be further developed to be able to work onboard a spacecraft. Running AI programmes in space has a number of benefits. Firstly, preprocessing data makes more effective use of the downlink channel because only ‘meaningful’ information is transmitted. Then, for systems being used to identify potentially threatening NEOs, observations can be made around the clock, whereas ground-based systems are limited to night-time use when the skies are dark.
Being generic, VIPER will be applicable to various forms of visual data, including multi-temporal datasets such as a time series of still images or video, or analysis of medical images such as X-rays. It may also be incorporated into Starion’s SmartDIG service, which is currently in development.
Further information
Contact details: Isabelle Roels, VP Marketing and Communications (i.roels@stariongroup.eu)
[1] https://www.sciencedirect.com/science/article/pii/S0048969723072121
[2] https://www.deloitte.com/global/en/issues/climate/the-catalytic-potential-of-artificial-intelligence-for-earth-observation.html
