Deep learning to enhance maritime situation awareness

EasyChair Preprint 891

Abstract

Maritime surveillance sensors like AIS (Automatic Identification System) and Radar provide useful information for decision-making support, which is of paramount importance for effective operations against maritime threats and illegal activities [1]. However, decision-making systems that trust solely on AIS information tend to fail in real situations because such information could be missing, inaccurate or even deceptive [2]. On the other hand, only Radar information is not enough to get a complete description of the maritime situational picture. This paper proposes a deep learning framework for vessel monitoring that examines a particular scenario where a deep learning solution can infer a navigation status based on the vessels trajectories, and thus to detect suspicious vessels activities. For this purpose, a dataset, named DeepMarine, has been specifically created by collecting data of AIS historical recordings. We demonstrate the performance of the developed deep learning framework for the proposed vessels activity classification, which can be ultimately used to report illegal activities.

Keyphrases: AIS, Radar, ResNet, deep learning, navigation status, ship trajectories, vessel activity

@booklet{EasyChair:891,
  author    = {Tomás Mantecón and David Casals and Juan Jose Navarro-Corcuera and Carlos Roberto Del-Blanco and Fernando Jaureguizar},
  title     = {Deep learning to enhance maritime situation awareness},
  doi       = {10.29007/7ldh},
  howpublished = {EasyChair Preprint 891},
  year      = {EasyChair, 2019}}