NHPL: Neural Collapse Guided Hierarchical Prototypical Learning for Radar HRRP Open Set Recognition

TY - JOUR

T1 - NHPL

T2 - Neural Collapse Guided Hierarchical Prototypical Learning for Radar HRRP Open Set Recognition

AU - Wang, Yanhua

AU - Liu, Yichen

AU - Yu, Binhua

AU - Wang, Weijia

AU - Zhang, Liang

AU - Zhang, Xin

AU - Li, Yang

N1 - Publisher Copyright: © 1965-2011 IEEE.

PY - 2026

Y1 - 2026

N2 - High-resolution range profile (HRRP) has emerged as a promising approach in real-time radar ground target recognition. To address real-world operational requirements, there is a need to develop open set recognition (OSR) methods that identify both known and unknown classes of targets. In this article, we propose a neural collapse-guided hierarchical prototypical learning method for HRRP OSR. The key aspect of prototypical learning is the construction of class-specific prototypes in the feature space as representatives. However, this is challenging in HRRP recognition due to the intrinsic aspect sensitivity, which results in variations in feature distribution and deviations in estimated prototypes. To tackle the aforementioned issues, we employ a hierarchical classification strategy that decomposes the OSR task into a multilevel recognition framework based on the hierarchical semantic taxonomy of classes. Within each level, discriminate features can be more effectively learned, thereby constraining variation. Furthermore, inspired by neural collapse theory, we estimate class prototypes using the weight vectors of the final layer of the trained network. This approach aligns prototypes closely with features of known classes while staying distant from features of unknown classes, thereby reducing prototype deviation. Extensive experiments on measured HRRPs demonstrate that the proposed method outperforms existing methods in terms of accuracy and robustness for recognizing unknown classes.

AB - High-resolution range profile (HRRP) has emerged as a promising approach in real-time radar ground target recognition. To address real-world operational requirements, there is a need to develop open set recognition (OSR) methods that identify both known and unknown classes of targets. In this article, we propose a neural collapse-guided hierarchical prototypical learning method for HRRP OSR. The key aspect of prototypical learning is the construction of class-specific prototypes in the feature space as representatives. However, this is challenging in HRRP recognition due to the intrinsic aspect sensitivity, which results in variations in feature distribution and deviations in estimated prototypes. To tackle the aforementioned issues, we employ a hierarchical classification strategy that decomposes the OSR task into a multilevel recognition framework based on the hierarchical semantic taxonomy of classes. Within each level, discriminate features can be more effectively learned, thereby constraining variation. Furthermore, inspired by neural collapse theory, we estimate class prototypes using the weight vectors of the final layer of the trained network. This approach aligns prototypes closely with features of known classes while staying distant from features of unknown classes, thereby reducing prototype deviation. Extensive experiments on measured HRRPs demonstrate that the proposed method outperforms existing methods in terms of accuracy and robustness for recognizing unknown classes.

KW - Hierarchical classification

KW - open set recognition (OSR)

KW - prototypical learning

KW - radar automatic target recognition (RATR)

UR - https://www.scopus.com/pages/publications/105023294582

U2 - 10.1109/TAES.2025.3637742

DO - 10.1109/TAES.2025.3637742

M3 - Article

AN - SCOPUS:105023294582

SN - 0018-9251

VL - 62

SP - 2350

EP - 2368

JO - IEEE Transactions on Aerospace and Electronic Systems

JF - IEEE Transactions on Aerospace and Electronic Systems

ER -