Ship Magnetic Signature Classification Using GRU-Based Recurrent Neural Networks
Magnetic signatures represent the magnetic field generated by a ship’s ferromagnetic components and provide valuable information for identifying vessels not only in naval operations, but also in civil passages. The topic of accurate modelling of these signatures is relevant to this day, b...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10947760/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Magnetic signatures represent the magnetic field generated by a ship’s ferromagnetic components and provide valuable information for identifying vessels not only in naval operations, but also in civil passages. The topic of accurate modelling of these signatures is relevant to this day, but also the complexity of the model necessary to accurately predict the ship’s magnetic field. This paper presents the implementation of a deep, recurrent neural network (RNN) designed for classification of compliance between the original magnetic signature of a ship and the one obtained from a model. Therefore, the quality of the model can be analyzed using a classifier during the modeling process. The necessity to introduce a tool for signature compliance classification arose during numerical modeling of a ship in Finite Element Method (FEM) environment as well as during reverse modeling based on data coming from measurements. Another application is the use of a shallow RNN for classifying ships by their size and type. A sufficient amount of data is rarely available and therefore data augmentation solution is necessary. The process of obtaining a large dataset of signals from a multi-dipole model and using an interpolation technique for generating training, validation and test data is comprehensively described. Methods used for selecting the best network structure and hyperparameter tuning using grid search and random search in order to achieve a satisfactory classification accuracy are thoroughly explained. Features, advantages and limitations of developed algorithms are derived strictly from the nature of neural networks. |
|---|---|
| ISSN: | 2169-3536 |