Contribution
Ship Model for Machine Learning Based Ship Signature Classification
* Presenting author
Abstract:
The aim of a new project is to classify ship types in hydroacoustic data using machine learning. In a first step, the waterborne sound measurements are analyzed with regard to occurring events. The detected events are then grouped into different classes. In the last step, the events of the ship class are classified more precisely and assigned to specific ship types. Here, ship-specific hydroacoustic signatures, which are influenced by the propulsion concepts and engines, for example, provide features for classifying the ship type. As publicly available labeled datasets are rare, a model dataset is generated in addition to real recorded data. For this purpose, a ship model is equipped with actuators for the hydroacoustic manipulation of ship signatures so that different defined ship signatures can be reproduced and recorded. Finally, the suitability of integrating the ship model data to train the network structures for recognizing real ships will be investigated. This paper gives a short overview of the project, especially the work package of data generation using a ship model.