Missing Modality Completion for Multi Frequency Radar HRRP Recognition Using GAN

High-resolution range profile (HRRP) is one of the commonly used methods in radar automatic target recognition (RATR). Recently, obtaining HRRP under different modalities, such as frequency and polarization, to improve the RATR performance has become an important development direction. However, in practical applications, sensor failures or interference may lead to modality loss during the model testing phase, resulting in decreased performance or even causing the model to fail to function properly. This paper primarily addresses the scenario of frequency modality loss and specifically proposes a modality completion and recognition network with generative adversarial network-embedded (MCRN-GAN) model. The model first generates and completes the missing modality features using the frequency feature transformation loop (FFTL) based on GAN. Subsequently, it fuses multi-modal features based on spatial and channel attention mechanisms to obtain a joint representation of dual-frequency features. Finally, a fully connected layer is used to complete the classification. Experiments on Ku and W frequency modality datasets demonstrates that the proposed method can effectively handle missing modality data after sufficient training without significant performance loss.