现代雷达2025，Vol.47Issue(10)：110-115,6.DOI:10.16592/j.cnki.1004-7859.2025052701
基于特征增强和特征压缩的模型在线更新方法
Online Model Update Method Based on Feature Augmentation and Compression
张琨 1李书玉 2徐文玲 3吴春晓 1杜亮 1程强1
作者信息
- 1. 南京电子技术研究所,江苏 南京 210039||雷达探测感知全国重点实验室,江苏 南京 210039
- 2. 解放军63612 部队,甘肃 敦煌 736200
- 3. 解放军93236 部队,北京 100085
- 折叠
摘要
Abstract
Radar automatic target recognition(RATR)and inverse synthetic aperture radar(ISAR)technologies play a crucial role in modern electronic warfare and target detection.ISAR enables high-resolution imaging of fast-moving targets in complex environ-ments,with wide applications in target recognition,tracking,stealth detection,and electronic countermeasures.As battlefield sce-narios become increasingly complex and the diversity of targets grows,traditional recognition systems struggle with limited adapta-bility and the problem of catastrophic forgetting,making them inadequate for continuous updates and accurate identification.To ad-dress these challenges,this paper proposes a continual learning framework with enhanced anti-forgetting capabilities.By integrating exemplar replay and feature augmentation strategies,the proposed method alleviates the performance degradation caused by the im-balance between new and old classes.In addition,residual fitting and knowledge distillation techniques are employed to capture in-ter-class variations and reduce model complexity,enabling a better balance between stability and plasticity.This approach demon-strates strong potential in improving the adaptability and efficiency of radar target recognition systems,providing an effective solu-tion for continual learning in dynamic and evolving battlefield environments.关键词
雷达自动目标识别/逆合成孔径雷达/持续学习/样本回放/知识蒸馏Key words
RATR/ISAR/continual learning/exemplar replay/knowledge distillation分类
电子信息工程引用本文复制引用
张琨,李书玉,徐文玲,吴春晓,杜亮,程强..基于特征增强和特征压缩的模型在线更新方法[J].现代雷达,2025,47(10):110-115,6.
