CROF-Net: A Robust Algorithm for Simulation-Measurement Cross-Domain Few-Shot HRRP Target Recognition

Aiming at limited measured HRRP samples of non-cooperative targets and severe distribution discrepancy between simulated and real echoes, this paper proposes CROF-Net, a robust cross-domain few-shot recognition framework for complex electromagnetic environments. It first corrects simulated data distribution via low-order moment matching and enhances few-shot measured samples for domain consistency. A 1D-Conformer backbone embedded with instance normalization and scattering-peak-aware modules extracts robust local and global features from distorted HRRP signals. A reliability-aware prototype calibration module achieves fine-grained class-level alignment. To handle attitude-induced intra-class multimodal distribution, we design an orthogonal-constrained sub-center ArcFace metric space to broaden classification boundaries and relieve negative transfer and catastrophic forgetting. Evaluated on SAMPLE and MTDSP datasets, CROF-Net achieves higher accuracy and stronger anti-interference capability than existing methods, presenting great research value and engineering application prospects.