FIRE-SAT

Starting point / motivation

Artificial Intelligence (AI) and the development of (constellations of) space crafts with small form factor have been two important technological paradigm shifts in recent times. Combining them and bringing AI on board of satellites and space probes has the potential to remove limitations due to data transmission bandwidth and time delay and thereby enable completely new space mission capabilities. Such novel AI-enabled space mission scenarios could, in particular, become powerful tools to contribute towards achieving the sustainable development goals (SDG) against the backdrop of the climate change and species loss crises by detecting and timely alerting the onset of environmental catastrophes such as wildfires, (oil) pollution, habitat destruction, etc.

At this point however, machine-learning-based processing of data on board of earth observation (EO) satellites is still very much in a research and developing stage, characterized by first demonstrations of processing capabilities and experimental satellites, including ESA’s Phi-SAT#1/2 and OPS-SAT missions, as well as, developments by “NewSpace” start-ups.

Contents and goals

The proposed project by Silicon Austria Labs and its partners OroraTech and Joanneum Research aims at realizing a proof-of-concept that explores feasibilities of an onboard-AI-enabled responsive EO mission.

Methods

In FIRE-SAT tailored state-of-the-art machine-learning methodologies are deployed to operational satellite sensors to process RGB imaging data onboard for a remote fire detection use case. The satellites for the in-flight experiment are ESA's OPS-SAT mission (sun-synchronous dawn/dusk orbit 3U CubeSat launched 2019), which provides an experimental platform for registered users to test new ideas in mission scenarios and operations, and a satellite of the “NewSpace” company OroraTech. In in-flight experiments, the satellites’ live camera images are analysed onboard for the presence of smoke plumes by suitably trained and FPGA/GPU-implemented resource-constrained convolutional neural networks.

In this regard the participation of the "NewSpace" company OroraTech – a provider of wildfire information services and a developer of a dedicated CubeSat constellation – as an international partner in the collaborative exploratory project, not only contributes in-depth use case & data expertise, but adds unique value to the project in terms of satellite availability and ecosystem beyond the OPS-SAT mission with a satellite that is optically compatible and complementary with respect to processing hardware (GPU vs. FPGA).

Expected results

The methodologies developed in the course of the project represent an approach that holistically investigates the feasibility of acquisition/generation of suitable data (training/validation set), machine learning modelling & experiment application concept and software development and the implementation on resource-constrained embedded hardware aspects of an EO mission enabled by onboard AI.

An inevitable prerequisite for the onboard geolocation of detected (fire) events or the onboard detection of changes in multi-temporal acquisitions is the precise geocoding of data using the spacecraft’s processing hardware.

FIRE-SAT thus includes work on high-performance, onboard geocoding capabilities by Joanneum Research. If the project delivers promising results, it is planned to continue the work in an R&D&I project with the use cases of onboard land-cover change detection.

Coordinator

Silicon Austria Labs GmbH

Project partner

• JOANNEUM RESEARCH Forschungsgesellschaft mbH
• Orbital Oracle Technologies GmbH

Silicon Austria Labs GmbH
Dr. Lothar Ratschbacher,
Alternberger Str. 69
A-4040 Linz