Hierarchical classification of ISAR sequences

Inverse synthetic aperture radar (ISAR) is a common radar imaging technique used to characterise and classify non-cooperative targets. A number of different classification approaches have been proposed including the traditional approach which utilises geometric features extracted from images of known targets and more recently deep learning approaches. Hierarchical classification is designed to deal with large scale classification problems when there is a large number of targets. In this paper, a number of convolutional neural networks are compared to determine the best approach for hierarchical ship classification. Recent work on transfer learning, recurrent neural networks, and hierarchical structures are explored with the goal to find the best network design in terms of computational efficiency and classification accuracy.