Frequency- and Angle-Domain RCS Extrapolation of Cluster Targets Based on Physics-Informed Stacking Framework

This paper presents a physics-informed stacking framework (PISF) for extrapolating the frequency-dependent radar cross section (RCS) of cluster targets. A stacked ensemble model is constructed by integrating machine learning (ML) and deep learning (DL) techniques to improve prediction accuracy in the frequency domain. A hybrid loss function is introduced, combining data-driven and physics-guided components to enhance model interpretability and physical consistency. The model is initially trained on synthetic data and subsequently fine-tuned using inductive transfer learning (ITL) with domain-adversarial adaptation on measured datasets of complex clustered targets, thereby improving generalization capability while ensuring computational efficiency. The proposed method achieves an RCS extrapolation root-mean-square error (RMSE) of less than 2 dB on simulated datasets. On measured data, it reduces the required sample volume by 20%–33% while maintaining high predictive robustness.