Deep models for instance segmentation of radar chirps for maritime surveillance

Details

Supervisors

De Vleeschouwer, Christophe ; Vandendorpe, Luc

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil électricien, à finalité spécialisée

Abstract

enLocating ships that disable their identification systems is an important challenge for maritime surveillance. One solution is to observe the radar signals that ships continue to emit, though this is difficult due to the presence of signals from multiple unknown emitters. This master thesis proposes an alternative approach to radar signal deinterleaving by analyzing received signals directly in the time–frequency domain. A synthetic dataset was created to evaluate the viability of this approach using deep learning models. The results show that this approach can effectively distinguish and isolate overlapping radar emissions, suggesting that time–frequency analysis is a promising direction for future work.