Neural network model of ship magnetic signature for different measurement depths

This paper presents the development of a model of a corvette-type ship’s magnetic signature using an artificial neural network (ANN). The capabilities of ANNs to learn complex relationships between the vessel’s characteristics and the magnetic field at different depths are proposed as an alternative to a multi-dipole model. A training dataset, consisting of signatures prepared in finite element method (FEM) environment Simulia Opera was constructed. A feedforward neural network was developed through a comparative analysis of different activation functions available in MATLAB’s Deep Learning Toolbox and the grid search method. Verification was performed using the leave-one-out cross-validation method (LOOCV). The model proved to be highly effective in predicting the magnetic signature for the northward direction in any measurement depth, with prospects to expand it to estimate other directions.