An open-set method for automatic recognition of low probability of intercept radar waveforms based on reciprocity points learning

Abstract

Deep learning-based methods for Low Probability of Intercept radar waveform recognition typically assume that the signal to be recognized belongs to a known and finite set of classes. However, in practical scenarios, the electromagnetic signal environment is open and there may be a large number of unknown signals, making such methods difficult to apply. To address this issue, a novel open-set recognition method based on reciprocal points is proposed. This approach uses a neural network to extract a high-dimensional time-frequency feature map of the signal, and measures the difference between the known and unknown signals by computing the distance between the feature vector and the reciprocal points. This allows the model to correctly identify known class signals while simultaneously detecting unknown signals. Experimental results show that the proposed method achieves open-set recognition of Low Probibability of Intercept radar signals. On test signals with signal-to-noise ratios ranging from 6 dB to 15 dB, the model achieves nearly 100% accuracy in identifying known class signals and more than 90% accuracy in detecting unknown signals.