Contribution
Short-Term Ship Trajectory Classification Based on AIS Data
* Presenting author
Abstract:
In the context of maritime port monitoring and threat identification, a novel approach is presented by passive Sound Navigation and Ranging (SONAR) methods in combination with machine learning algorithms. Vessels passing a SONAR surveillance setup emit different sound profiles and can thus be detected, localized, classified, and ideally, uniquely identified. To assist in the development of the required algorithms and acquisition of data, this work proposes a method for classification into different ship types based on (position-only) information from the Automatic Identification System (AIS). Vessel trajectories are obtained and divided into short-term path segments, which are labeled with the correct vessel type. These segments are then filtered and used to train a Convolutional Neural Network (CNN). The output layer of the CNN is a soft classification between four basic ship types that show differences in their behaviour. A good accuracy is achieved for vessel types which are well represented in the training data. The method can subsequently be applied to similar trajectory segments obtained from different sources, such as SONAR measurements.