A Robust Algorithm for Simulation‐Measured Cross‐Domain Few‐Shot HRRP Target Recognition

ABSTRACT

Aiming at the scarcity of labelled measured high‐resolution range profile (HRRP) samples for 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 adapted to complex electromagnetic environments. First, low‐order moment matching is adopted to calibrate the distribution of simulated data and measured samples are enhanced to improve domain consistency. A 1D‐conformer backbone integrated with instance normalisation and scattering‐peak‐aware modules is constructed to extract robust local and global features from distorted HRRP signals. To address the intra‐class multimodal distribution caused by target attitudes, we establish an orthogonal‐constrained sub‐centre ArcFace metric space, which widens classification boundaries and mitigates negative transfer as well as catastrophic forgetting. Experimental results on the SAMPLE and MTDSP datasets demonstrate that CROF‐Net outperforms in recognition accuracy and anti‐interference capability and it possesses promising research value and engineering application prospects.