Ships Classification using Neural Network based on Radar Scattering

Abstract

An ability to identify maritime ships as well as their different kinds is an essential part of modern maritime surveillance, safety, as well as security. The automatic classification of ships in Electro-Optical as well as Inverse Synthetic Aperture Radar (ISAR) images is a considerable challenge. In the present work, an application of deep convolutional neural networks in EO images along with the sequential algorithms for ISAR images has been proposed. Various CNN models have been configured and evaluated in order to seek the best hyper-parameters and the most suitable results are found by using transfer learning at different layers for EO images. The experiments conducted illustrate a success rate of more than 92% for this approach, also overpowers the traditional CNN-based techniques. A sequential algorithm composed of neural networks for category definition and rule-based classification for class identification is implemented for ISAR images. The dataset used for training, validation, and testing is simulated for imagery as well  as for real imagery.