Contribution
Classification of Vessel Types by Means of Machine Learning
* Presenting author
Abstract:
Monitoring and identification of ships and underwater objects are essential tasks for securing maritime infrastructures such as ports and shipping routes. Moving vessels produce characteristic sound waves based on ship-specific parameters such as propeller type and speed, which can be detected by passive SONAR systems. Currently, experienced human SONAR operators classify these sounds, often with impressive success rates, but automated decision-making approaches can assist new operators or even allow autonomous systems to perform classification independently.In our study, we developed a machine learning-based vessel classification system that uses underwater acoustic recordings to identify vessels and determine their class. In this process, the recordings are pre-processed with time-frequency analysis to extract features. These features are used as input to train and evaluate a convolutional neural network (CNN) specialized for classification. To maximize performance, the CNN undergoes several training cycles with different configurations and structures. Subsequently, the preprocessing algorithms and the trained CNN are implemented in a real-time framework.