Client
SPACEBEL, MULTITEL, ALX Systems, and ULiège are involved, with DELTATEC serving as the project coordinator.
Project
The IADAS project, accredited by the Skywin cluster as part of the Marshall Plan for the Walloon Region, aims to exploit the technological and economic convergence between the fields of UAVs and NewSpace satellites, resulting from the NewSpace revolution and UAV autonomy.
The main objective is to equip autonomous drones and NewSpace satellites with significant data processing capabilities thanks to artificial intelligence. This initiative will enable drones to perform a variety of functions for complex missions, as well as real-time processing of images generated by increasingly powerful on-board cameras, resulting in significant data growth.
Challenge(s)
The first challenge was to adapt AI algorithms for embedded platforms, reducing the size of neural networks while preserving their performance, and to develop multispectral and hyperspectral image processing for Earth observation (drones and satellites).
The second challenge was to design embedded platforms for UAVs and NewSpace, characterized by their small size, light weight, power, efficiency and thermal stability.
Solution
DELTATEC has developed an algorithm reduction workflow compatible with TensorFlow and PyTorch, incorporating custom and catalog algorithms, as well as datasets. To address the challenge of obtaining sufficient training data for neural networks, especially in drone multispectral imaging, DELTATEC created realistic synthetic data for applications like tree detection in dense forests, utilizing techniques such as frequential neural networks and style transfers.
The project's ultimate goal is to create an embedded and efficient platform capable of hosting AI algorithms. DELTATEC extensively studied various AI platforms and processors designed for edge devices, including HAILO, JETSON, and FPGAs.