Deep learning for maritime traffic surveillance
Within SESAME project, we have obtained promising preliminary results for the analysis, modeling and recognition of behaviours of vessels from AIS data streams using deep learning models. More specifically, variational Recurrent Neural Nets allow us for fully exploiting large-scale noisy and irregularly-sampled AIS datastreams to learn computationally-efficient representations of complex spatio-temporal patterns. Results on real AIS datasets demonstate the relevant of the proposed approach as illustrated below. These results are presented in a paper recently published at the IEEE Int. Conference on Data Science and Advanced Analytics, IEEE DSAA’2018. The preprint can be found online (link).
Reference: Nguyen et al., A Multi-task Deep Learning Architecture for Maritime Surveillance using AIS Data Streams, IEEE Int. on Data Science and Advanced Analytics, IEEE DSAA’2018, Torino, Oct. 2018 (link).
