.png){kind=icon}
Return
Selective Structural Completion Diffusion Model for HRRP Open-Set Recognition
.png){kind=icon}
PRE
AI
DOI：10.1109/taes.2026.3725223.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
High-Resolution Range Profile (HRRP) based recognition is pivotal for radar sensing, yet the presence of unknown targets in operational environments necessitates effective Open-Set Recognition (OSR). While reconstruction-based generative approaches have shown promise, conventional holistic reconstruction often fails to capture the intrinsic structural dependencies among sparse scattering centers in HRRPs. This is largely because the self-mapping objective can be satisfied via trivial shortcuts, allowing unknown targets to be reconstructed with deceptively low errors. Moreover, the holistic reconstruction loss tends to dilute critical structural deviations by averaging errors over the entire profile, which implicitly down-weights sparse yet informative scattering regions. To address these issues, we propose the Selective Structural Completion Diffusion Model (SCDM), which reframes HRRP OSR as a masked structural completion task. Our framework first employs a Structural Saliency Mask Module (SSM) to selectively mask salient segments based on attention saliency and a scattering intensity prior, forcing the model to focus on structurally critical regions. With the SSM as local guidance, the Structure-Prior Guided Diffusion Module (SGDM) incorporates global structural context to ensure faithful completion of scattering details. These modules are unified within a Bayesian joint optimization framework, where the derived conditional reconstruction residual serves as a reliable criterion for open-set rejection. Extensive experiments on a measured HRRP dataset demonstrate the effectiveness of SCDM in challenging OSR scenarios.
Keywords:
Radar automatic target recognition
high-resolution range profile
open-set recognition
diffusion models
deep learning
Journal
IF:
5.7
Papers:
797
Citations:
2.4W
Organization
View more organizations
Cited Papers
No cited papers available