Identity consistency association based on Cauchy-kernel semiparametric model optimization

Identity association under multi-source intermittent observations is a fundamental challenge in space object cataloging, space situational awareness, and complex measurement-data processing. To address the distortion of tracklet similarity caused by short arcs, weak observability, compounded errors, and interference from neighboring objects, a novel identity association method based on Cauchy-kernel semiparametric model optimization is proposed. First, through data modeling, identity association between intermittent measurement tracklets is formulated as a same-origin hypothesis testing problem based on measurable invariants. Second, heavy-tailed anomalous residuals are suppressed using a Cauchy-kernel loss, while systematic errors shared by measurement sources are compensated for using a semiparametric bias model, thereby yielding debiased measurable invariants. Then, the proportions of error components attributable to systematic bias, heavy-tailed anomalies, and quality degradation are adaptively estimated and propagated into an effective covariance to characterize the uncertainty of the measurable invariants. Furthermore, a profile-aware identity-consistency statistic is constructed to enhance the association capability for neighboring objects with different identities. Experimental results demonstrate that, compared with Euclidean-distance association, Mahalanobis-distance association, single-model gated association, maximum-correntropy interacting multiple-model methods, and kinematic-feature matching, the proposed method achieves superior identity-association performance.