Feature Extraction and Classification of Simulated Monostatic Acoustic Echoes from Spherical Targets of Various Materials Using Convolutional Neural Networks

Active sonar target classification remains an ongoing area of research due to the unique challenges associated with problem (unknown parameters, dynamic oceanic environment, different scattering mechanisms, etc.). Many feature extraction and techniques have been proposed, but there a need relate explain classifier results in physical domain. This work examines convolutional neural networks trained on simulated data known ground truth projected onto two time-frequency representations (spectrograms scalograms). The classifiers were discriminate material type, geometry, internal fluid filling, while hyperparameters tuned task using Bayesian optimization. examined explainable artificial intelligence technique, gradient-weighted class activation mapping, uncover informative features used discrimination. analysis resulted visual that allowed CNN choices be related It was found scalogram representation provided negligible accuracy increase compared spectrograms. Networks between geometries highest accuracy, lowest accuracy.