SPACE APPLICATIONS SERVICES S.A./N.V. (BE)
The AIOPEN project will combine and extend the existing platform ASB (Automated Service Builder), EOPEN (Open Interoperable Platform for Unified Access & Analysis of EO Data) and EOEPCA (EO Exploitation Platform Common Architecture) with new and innovative services based on operationally mature AI/ML software capabilities to provide a platform that supports the end-to-end AI model development lifecycle.
The platform, hosted in the ONDA DIAS, owned by Serco, will be capable to distribute processing tasks in remote environments. The result will be a public commercial service offering a dynamic pricing structure with a remuneration policy for contributors to the platform content (with models or data) or for performing activities such as labelling data and training AI models. AIOPEN will bring together:
The service will allow versioning, sharing and customising the various AI/ML resources and provide the tools to integrate/exploit the AI/ML models in new applications, for example exposing these ones via programming interfaces for running predictions.
Based on the Automated Service Builder (ASB) and EOPEN, developed by Space Applications Services, AIOPEN will allow importing custom processes, creating workflows and executing them in a distributed environment to deliver services on user customisable dashboards. Components from ESA/ESRIN EO Exploitation Platform Common Architecture (EOEPCA) project, led by Telespazio, will be included to bring interactive development, cataloguing and sharing capabilities. Popular open-source software will be integrated to include AI capabilities required by the AI model development lifecycle such as the ability to version and store model training projects, publish available models and datasets, annotate raw data, further train models and use models to do predictions. Key operations such as model training and predictions generation will also be available through programming interfaces.
In the course of the project, two showcases related to the Space for a Green Future theme will be implemented using the platform in order to demonstrate and validate its AI capabilities: Forest cover monitoring, proposed by KP Labs, and Urban Change Detection with Transformer Architecture, proposed by IT4Innovations (VSB —Technical University of Ostrava).
Forest cover monitoring, including deforestation tracking, can be achieved via segmentation and comparison of segmentation masks. This approach enables the usage of standard, well-known, favourably supervised, architectures like U-Nets for image segmentation. Moreover, images taken at any time interval can be subjected to this type of deforestation analysis. The deforestation tracking will use segmentation as an intermediate step. The resulting model will offer services like generating forest coverage masks, deforestation masks (via subtraction of forest masks), and quantification of deforestation (% of forests that were lost in the given time interval).
In the second showcase proposed by IT4Innovations we use EO data and Deep Neural Networks (DNNs) to detect (urban) related changes on the Earth’s surface to construct a digital twin of Earth’s (urban) changes. Detection of how urban areas, cities, infrastructure, and urban sprawl change over time helps to understand the dynamics of how the environment is impacted, to identify new (illegal) settlements, and to extrapolate trends for future planning. The showcase uses modern transformer based architectures to demonstrate the versatility and performance of DNNs for urban change detection.