Unknown radar waveform recognition system via triplet convolution network and support vector machine

Abstract

We propose an unknown radar waveform recognition system for identifying unknown radar waveforms and classifying known radar waveforms simultaneously, which can be summarized into three stages: a training stage that a triplet convolution network is trained to capture features and a support vector machine is leveraged to classify signals with the assistance of known radar waveforms, a pseudo-testing stage that a proper threshold is chosen to set a boundary between known and unknown radar waveforms, and a testing stage that new signal types are distinguished from conventional types by loading models trained in the training stage and the threshold chosen in the pseudo-testing stage. Furthermore, we need to derive the corresponding central feature vector of each known class in the training stage so as to prepare for computing Pearson correlation coefficient in the pseudo-testing stage and the testing stage. Experiments demonstrate the effectiveness of the proposed system on classifying known signals and identifying unknown signals. In the context of adaptive electronic warfare, this research accomplishes identification of radar signal waveforms including unknown signals for the first time, laying a solid foundation for the progress of electronic reconnaissance system.