Organizations: Key Laboratory of Information Science of Electromagnetic Waves (MOE), and the Discipline and Technology Center of Microwave Vision Intelligent Sensing, Fudan University, Shanghai 200433, China
High-resolution range profiles (HRRPs) provide compact radar target representations but exhibit strong aspect sensitivity, complicating class-incremental recognition when historical data are unavailable. Changes in viewing aspect shift scattering-peak locations and amplitudes, while reflector groups may produce responses resembling ships. Existing dynamic expansion methods can also incur substantial parameter and memory overhead. To address these challenges, we propose P-SER, a Physical Consistency-Constrained Scattering Expandable Representation framework for exemplar-free class-incremental HRRP recognition. P-SER reconstructs dominant scattering points from complex-valued sum- and azimuth-difference-channel HRRPs and observation geometry. The reconstructed spatial structure, target extent, and energy distribution are used to derive class-level scattering prototypes and sample-level saliency priors, providing physically grounded constraints for feature learning and old-class knowledge retention. A primitive scatterer dictionary further characterizes local strong-scattering mechanisms and calibrates uncertain ship--reflector predictions at inference. Meanwhile, lightweight residual adapters are incrementally introduced into a shared HRRP backbone to learn new classes with limited model growth. Experiments on simulated and measured HRRP datasets show that P-SER delivers competitive recognition performance without retaining old-class exemplars, while maintaining physical consistency, interference robustness, knowledge retention, and model scalability.
Figures & tables
Fig. 1: Illustration of continual class-incremental HRRP observation and its physical scattering response. Target classes expand progressively, while the HRRP imaging mechanism remains intrinsically sensitive to viewing aspect. Dominant responses are formed by the range projection of physical scattering centers, motivating the use of stable physical properties for incremental recognition.
Fig. 2: Key challenges in class-incremental HRRP recognition and the corresponding solutions in P-SER.
Fig. 3: Overall framework of the proposed P-SER. The framework consists of three complementary components: physical-consistency scattering prototype constraint (PCSP), PSD-guided scattering-component inference calibration (PSD-SCIC), and lightweight residual adapter expansion (LRAE).
Fig. 4: Overview of PCSP, which reconstructs physically consistent scattering points from HRRP observations and uses the resulting class-level prototypes and sample-level saliency priors to constrain incremental recognition.
Fig. 5: Representative samples from the simulation and measured datasets: (a) HRRP profiles of the ten simulated target classes, including ships and reflector groups; and (b) measured target images with their corresponding HRRP profiles.
Dataset
Target type
Length & width (m)
Training
Test
Simulation dataset
Bulk carrier
(152.14, 16.07)
3240
810
Container
(139.37, 23.37)
3240
810
Passenger
(178.82, 26.01)
3240
810
General cargo
(121.50, 21.72)
3240
810
Dredger
(79.52, 15.42)
3240
810
TABLE I: Simulation and Measured HRRP Datasets
Fig. 6: Accuracy comparison of different incremental learning methods across incremental stages on the self-constructed simulation dataset.
Method
Exemplars per class
8-stage
4-stage
2-stage
Params. (M)
AB (%)
Aˉ (%)
Params. (M)
AB (%)
Aˉ (%)
Params. (M)
AB (%)
Aˉ (%)
LwF [ 29 ]
0
11.17
47.89
61.50
11.17
51.29
66.15
11.17
55.23
68.93
iCaRL [ 38 ]
150
11.18
43.28
62.42
11.18
48.45
64.57
11.18
49.11
65.69
PODNet [ 9 ]
150
11.18
56.34
69.01
11.18
59.03
72.15
11.18
60.05
72.18
DER [ 50 ]
150
89.39
60.41
73.12
44.71
63.37
77.98
22.37
69.41
78.45
VAE+DER [ 10 , 50 ]
150
89.44
65.55
76.69
44.74
69.97
80.91
22.40
71.06
83.03
TABLE II: Performance Comparison on the Self-Constructed Simulation Dataset
Fig. 7: Final-round confusion matrices of P-SER. The top row reports the simulation results under the (a) 8-stage, (b) 4-stage, and (c) 2-stage settings; the bottom row reports the measured results under the (d) 4-stage and (e) 2-stage settings.
Fig. 8: Accuracy comparison of different incremental learning methods across incremental stages on the measured dataset.
Method
Exemplars per class
4-stage
2-stage
Params. (M)
AB (%)
Aˉ (%)
Params. (M)
AB (%)
Aˉ (%)
LwF [ 29 ]
0
11.17
43.75
63.25
11.17
51.63
67.44
iCaRL [ 38 ]
150
11.18
53.14
65.47
11.18
55.29
68.07
PODNet [ 9 ]
150
11.18
48.62
63.81
11.18
52.81
64.90
DER [ 50 ]
150
44.71
60.47
69.32
22.37
65.47
72.12
VAE+DER [ 10 , 50 ]
150
44.74
61.72
71.34
22.40
64.38
74.77
TABLE III: Performance Comparison on the Measured Dataset
Setting
Component
Recognition accuracy (%)
Avg.
PCSP
PSD-SCIC
LRAE
Base
Stage-1
Stage-2
Stage-3
Stage-4
Stage-5
Stage-6
Stage-7
Stage-8
Ablation Exp-1
×
×
×
83.44
67.36
63.84
61.11
57.47
55.82
49.34
47.93
46.59
59.21
Ablation Exp-2
✓
×
×
96.29
89.37
87.24
84.68
80.15
77.92
73.46
67.81
65.29
80.25
Ablation Exp-3
✓
✓
×
96.29
89.37
85.76
83.18
82.64
78.31
75.87
74.22
72.59
82.03
Ablation Exp-4
✓
×
✓
96.29
93.72
87.35
85.72
84.19
81.63
77.84
75.27
74.51
84.06
Ablation Exp-5
×
✓
×
83.44
67.36
73.28
69.91
65.43
59.75
55.12
53.86
50.34
64.28
TABLE IV: Ablation study of the proposed components.
Fig. 9: Target reconstruction results for representative observations from the simulation test set. The dots denote the retained dominant scattering points, and the orange rectangles indicate the reconstructed target extents in the range and cross-range directions.
Target
GT (L,W) (m)
Reconstructed (L,W) (m)
ϵL (%)
ϵW (%)
Bulk carrier
(152.14, 16.07)
(149.38, 18.62)
1.81
15.87
Container
(139.37, 23.37)
(141.66, 25.97)
1.64
11.13
Passenger
(178.82, 26.01)
(186.59, 29.50)
4.35
13.42
General cargo
(121.50, 21.72)
(117.67, 23.09)
3.15
6.31
Dredger
(79.52, 15.42)
(83.25, 17.84)
4.69
15.69
Fishing
(46.85, 8.60)
(50.13, 11.81)
7.00
37.33
TABLE V: Full-Test-Set Reconstruction Statistics on the Simulation Dataset
Fig. 10: Hyperparameter sensitivity on the measured 4-stage protocol: final-round accuracy with respect to (a) λsp and λp , and (b) τp and μsc .
Few-shot high-resolution range profile (HRRP) recognition is challenged by strong aspect-dependent variation and sparse novel-class support. We propose polarimetric scattering-structure contrastive learning (PSS-CL), which supplies an episodic HRRP encoder with a structured physical training view. Dominant local responses are represented by their soft affinities to canonical polarimetric mechanisms, normalized range locations, and energy contributions. A symmetric cross-view objective transfers this scattering structure to the learned HRRP representation, while episodic prototype learning supports novel-class inference without the physical branch. We conduct Random-aspect and Cross-aspect experiments on a six-class measured dataset. In Cross-aspect 1-, 3-, and 5-shot recognition, PSS-CL achieves 81.73%, 86.92%, and 90.98% mean class accuracy (MCA), respectively, outperforming the best competing results by 5.62%, 3.09%, and 3.56%.
We revisit High-Resolution Range Profile (HRRP) classification with aspect-angle conditioning. While prior work often assumes that aspect-angle information is incomplete during training or unavailable at inference, we study a setting where angles are available for all training samples and explicitly provided to the classifier. Using three datasets and a broad range of conditioning strategies and model architectures, we show that both single-profile and sequential classifiers benefit consistently from aspect-angle awareness, with an average accuracy gain of about 7% and improvements of up to 10%, depending on the model and dataset. In practice, aspect angles are not directly measured and must be estimated. We show that a causal Kalman filter can estimate them online with a median error of 5{\textdegree}, and that training and inference with estimated angles preserves most of the gains, supporting the proposed approach in realistic conditions.
Edwyn Brient, Santiago Velasco-Forero, Rami Kassab
CMM · ARC, Thales Land and Air Systems Limours, France
High-resolution range profile (HRRP)-based radar automatic target recognition suffers from severe performance degradation in composite jamming environments. Active jamming introduces suppression- and deception-related components into the received range profile. After pulse compression, these components are coupled with target echoes in the HRRP domain, making target-related scattering peaks difficult to distinguish and weakening feature separability. To address this problem, this paper proposes JointHRRP-Net, a unified framework for joint target-jamming recognition. A statistically constrained decoupling module is first developed to generate target-dominant and jamming-dominant latent branches from the mixed HRRP representation. Correlation-guided statistical constraints are imposed to suppress redundant cross-branch information and alleviate target-jamming feature entanglement. A multi-scale temporal encoding module is then designed to model local scattering structures and long-range range-cell dependencies, followed by a dual-expert decision module for single-label target classification and multi-label jamming classification. Experiments under diverse signal-to-jamming ratio (SJR) and signal-to-noise ratio (SNR) levels demonstrate that JointHRRP-Net outperforms representative baseline methods in both target recognition and composite jamming recognition. Open-set evaluation further shows that the learned target representation remains discriminative for unknown-target rejection. These results demonstrate the effectiveness and robustness of JointHRRP-Net in composite jamming scenarios.
Yunfei Zhao, Mei Liu, Shuowei Liu +2
College of Electronic Science and Technology, National University of Defense Technology, Changsha 410073, China