| KEYWORD |
Area Engineering
Reliability Evaluation of Convolutional Neural Networks for Multispectral Images Segmentation on Earth Observation Microsatellites
Thesis in external company
keywords ARGOTEC, ARTIFICIAL INTELLIGENCE, EDGE COMPUTING, EDGE-AI, IRIDE, LOW EARTH ORBIT, MICROSATELLITE
Reference persons ANNACHIARA RUOSPO, EDGAR ERNESTO SANCHEZ SANCHEZ
External reference persons Lorenzo Sarti; Niccolò Battezzati
Research Groups DAUIN - GR-05 - ELECTRONIC CAD and RELIABILITY GROUP - CAD
Thesis type SOFTWARE SPERIMENTALE
Description In recent years, the Microsatellites market has grown rapidly thanks to their compact form factor, which allows for the reduction of launch and development costs typical of the space sector. In this context, Argotec, a leading company in the development of Microsats for both Deep Space and Low Earth Orbit (LEO) missions, participates in the Italian IRIDE program for Earth Observation with a constellation of satellites equipped with high resolution multispectral payloads. Argotec's technological platform, HAWK, will be able to count on a Payload Data Processor (PDP) for the implementation of AI algorithms on-the-edge.
Argotec proposes a thesis project for the development of a tool/framework for evaluating the reliability of convolutional neural networks for segmentation tasks on multispectral images on board these Microsats. Therefore, starting from the identification of relevant use cases in the context of the IRIDE program, a method will be defined for evaluating the reliability of the models selected for these use cases. The framework development phase and validation will follow directly on the Payload Data Processor selected for the constellation. Given the growing number of increasingly aware and autonomous satellite platforms, this thesis will contribute to the possibility of accurately evaluating the reliability of the models used on board, playing a key role in the success of the mission and the duration of the satellite in orbit.
Required skills • Theory of Machine Learning and Deep Learning
• Convolutional Neural Networks
• Python and C
• Linux based Operating Systems
• Notions on Reliability of AI Systems
Deadline 11/12/2024
PROPONI LA TUA CANDIDATURA