Kolmogorov–Arnold Networks for High-Resolution Range Profile Recognition In Radar Applications

Kejun Ye; Yanhua Wang; Rola Saad

doi:10.1109/IGARSS55030.2025.11244029

Kolmogorov–Arnold Networks for High-Resolution Range Profile Recognition In Radar Applications

Kejun Ye
, Yanhua Wang
, Rola Saad

University of Sheffield
Beijing Institute of Technology

Research output: Contribution to journal › Conference article › peer-review

Abstract

Conventional CNNs are used as the mainstream neural networks for recognizing HRRP signals, however, CNNs requires a large number of training samples. In this paper, a lightweight model combining Kolmogorov-Arnold network (KAN) and convolutional neural network (CNN) for high-resolution distance image (HRRP) target recognition is proposed. This method improves the classification accuracy in noisy environments and small sample scenarios through modular decomposition and B-Spline activation function. Experimental results demonstrate that the novel combined method exhibits superior generalization ability under small sample conditions and achieves good accuracy in the test set, providing an efficient and robust solution for HRRP target recognition.

Original language	English
Pages (from-to)	7365-7369
Number of pages	5
Journal	International Geoscience and Remote Sensing Symposium (IGARSS)
DOIs	https://doi.org/10.1109/IGARSS55030.2025.11244029
Publication status	Published - 2025
Externally published	Yes
Event	2025 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2025 - Brisbane, Australia Duration: 3 Aug 2025 → 8 Aug 2025

Keywords

HRRP
Kolmogorov-Arnold networks （ KAN）
convolutional neural networks（CNN）
radar signal processing
target recognition

Access to Document

10.1109/IGARSS55030.2025.11244029

Cite this

@article{cdab8a78cc1e4d878dc13ebb8908b184,

title = "Kolmogorov–Arnold Networks for High-Resolution Range Profile Recognition In Radar Applications",

abstract = "Conventional CNNs are used as the mainstream neural networks for recognizing HRRP signals, however, CNNs requires a large number of training samples. In this paper, a lightweight model combining Kolmogorov-Arnold network (KAN) and convolutional neural network (CNN) for high-resolution distance image (HRRP) target recognition is proposed. This method improves the classification accuracy in noisy environments and small sample scenarios through modular decomposition and B-Spline activation function. Experimental results demonstrate that the novel combined method exhibits superior generalization ability under small sample conditions and achieves good accuracy in the test set, providing an efficient and robust solution for HRRP target recognition.",

keywords = "HRRP, Kolmogorov-Arnold networks （ KAN）, convolutional neural networks（CNN）, radar signal processing, target recognition",

author = "Kejun Ye and Yanhua Wang and Rola Saad",

note = "Publisher Copyright: {\textcopyright}2025 IEEE.; 2025 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2025 ; Conference date: 03-08-2025 Through 08-08-2025",

year = "2025",

doi = "10.1109/IGARSS55030.2025.11244029",

language = "English",

pages = "7365--7369",

journal = "International Geoscience and Remote Sensing Symposium (IGARSS)",

issn = "2153-6996",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - JOUR

T1 - Kolmogorov–Arnold Networks for High-Resolution Range Profile Recognition In Radar Applications

AU - Ye, Kejun

AU - Wang, Yanhua

AU - Saad, Rola

PY - 2025

Y1 - 2025

N2 - Conventional CNNs are used as the mainstream neural networks for recognizing HRRP signals, however, CNNs requires a large number of training samples. In this paper, a lightweight model combining Kolmogorov-Arnold network (KAN) and convolutional neural network (CNN) for high-resolution distance image (HRRP) target recognition is proposed. This method improves the classification accuracy in noisy environments and small sample scenarios through modular decomposition and B-Spline activation function. Experimental results demonstrate that the novel combined method exhibits superior generalization ability under small sample conditions and achieves good accuracy in the test set, providing an efficient and robust solution for HRRP target recognition.

AB - Conventional CNNs are used as the mainstream neural networks for recognizing HRRP signals, however, CNNs requires a large number of training samples. In this paper, a lightweight model combining Kolmogorov-Arnold network (KAN) and convolutional neural network (CNN) for high-resolution distance image (HRRP) target recognition is proposed. This method improves the classification accuracy in noisy environments and small sample scenarios through modular decomposition and B-Spline activation function. Experimental results demonstrate that the novel combined method exhibits superior generalization ability under small sample conditions and achieves good accuracy in the test set, providing an efficient and robust solution for HRRP target recognition.

KW - HRRP

KW - Kolmogorov-Arnold networks （ KAN）

KW - convolutional neural networks（CNN）

KW - radar signal processing

KW - target recognition

UR - https://www.scopus.com/pages/publications/105034109133

U2 - 10.1109/IGARSS55030.2025.11244029

DO - 10.1109/IGARSS55030.2025.11244029

M3 - Conference article

AN - SCOPUS:105034109133

SN - 2153-6996

SP - 7365

EP - 7369

JO - International Geoscience and Remote Sensing Symposium (IGARSS)

JF - International Geoscience and Remote Sensing Symposium (IGARSS)

T2 - 2025 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2025

Y2 - 3 August 2025 through 8 August 2025

ER -

Kolmogorov–Arnold Networks for High-Resolution Range Profile Recognition In Radar Applications

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this