HRRPSeqNet: Open-Set Recognition of Space Target Motions Using HRRP Sequences

Identifying and analyzing the motion states of space targets helps to ensure space safety. Given the possible occurrence of unknown motions not covered by the dataset in observations, space target motion recognition can be considered an open-set recognition (OSR) task. Compared to other radar broadband data, high-resolution range profile (HRRP) sequences are more readily acquired and contain rich target dynamic information, making them well-suited for motion analysis. In this article, we introduce the HRRP sequence network (HRRPSeqNet), a lightweight OSR model. This model uses the partial inception module to extract feature efficiently and the label-guided module to promote feature clustering. To cope with geometric modeling errors, we incorporate HRRP structural features and jointly model them with HRRP sequences in the label-guided module. To effectively reject unknown classes, we propose the KL-divergence weighted open-set inference algorithm, which adjusts inference results of HRRPSeqNet. Simulation and real data experiments validate the effectiveness of the proposed method.