Zero-Shot Deception Jamming Recognition Based on Multipulse Integrated Spectrum Reconstruction

In electronic countermeasure scenarios, the recognition between physical target (PT) and active false target (FT) is a prerequisite for the radar to effectively measure target parameters and realize target tracking. However, the existing recognition methods-based on artificial intelligence (AI) suffer from weak noise immunity and excessive dependence on unrealistic training samples, hindering their adaptation to highly dynamic electronic countermeasures environments. Therefore, we propose a zero-shot, unsupervised autoencoder (AE)-based framework that leverages reconstruction loss differential (RLD) to discriminate between PT and FT. Firstly, the spectrum mismatch characteristics caused by phase superposition effects in signals with multiple scattering centers (SCs) are analyzed, thereby demonstrating the diversity and consistency exhibited by the multi-pulse integrated spectrum (IS) features of PT and FT in Netted radar (NR) systems. Then, to achieve zero-shot sampling on PT and FT, an AE comprising one-dimensional convolutional neural network (1DCNN) is designed for NR to reconstruct the IS features after cross correlation processing, and the discrimination between PT and FT is realized by using the distribution law of the RLD. Simulation and measured experimental results show that the proposed method has outstanding performance in low signal-to-noise ratio (SNR), zero-shot, and dynamic scenario cases, achieving a recognition accuracy exceeding 93% on measured data. The proposed method can be used in tandem with the data-level fusion based method to further eliminate the false alarm probability and improve the reliability of measurements.