Daniel Brooks is a PhD student at LIP6 (MLIA team) and has just defended his thesis on July 3rd. His work focuses on the classification of structured time series, particularly in the form of radar micro-Doppler signals from non-cooperative UAVs.
We first focus on the rich internal structure of radar micro-Doppler signals, which provides a wide variety of physical representations, which in turn allow the development of various learning models, in particular convolutional neural networks, or on SPD matrices. We then show how to adapt some models to the internal structure of the data through the information geometry, and propose various improvements to these models. Finally, we develop a global classification pipeline, involving all representations and associated learning models, and present the results on different real and synthetic databases.