A Recognition Model for MFR Operating Mode Based on Bayesian Inference and Deep Learning

Recognition of Multi-Function Radar (MFR) operating mode is critical for electronic reconnaissance. The diversity of MFR emission waveforms and complex transition patterns result in an incomplete MFR waveform sample library for reconnaissance. In complex electromagnetic environments, missing pulses in deinterleaved MFR pulse streams cause fragmented time-frequency features. These factors make MFR mode recognition a significant challenge in electronic reconnaissance. This paper proposes a recognition model based on Bayesian inference and deep learning. First, a Bayesian-based radar operating mode probability model is established, converting mode recognition into posterior predictive distribution inference, improving recognition for unknown signal waveform. The attention mechanism and Depthwise separable convolution extract global and local temporal features, enhancing the ability of the model to distinguish pulses from different modes. Simulation results show that the model demonstrates strong effectiveness and robustness.