Convolutional neural networks for radar emitter classification

In this paper, an application of convolutional neural networks (CNN) for rapid and accurate classification of electronic warfare emitters is investigated; a large data set with 58 separate emitter sources is used for training and testing. Data preprocessing creates 3-dimensional images with a feature space composed of pulse width (PW), radio frequency (RF), and pulse repetition interval (PRI), referenced with respect to time of arrival (TOA). The image representation has proven to be the most effective, consistently producing classification accuracies approaching 98.7%. This study, which evaluates emitter-by-emitter classification, appears to be a novel approach, based on a survey of current literature; previous work citing the use of CNNs in this domain has been limited to radar waveform recognition vice pulse-based specific emitter identification.