| 引用本文： |
-
彭鹏,曹帅,贾勇,等.结合特征分布优化的辐射源类增量识别方法[J].电讯技术，2026，（4）：629 - 636. [点击复制]
- PENG Peng,CAO Shuai,JIA Yong,et al.Incremental Identification of Radiation Source Classes Combined with Feature Distribution Optimization[J].，2026，（4）：629 - 636. [点击复制]
|
|
| 本文已被：浏览 18次 下载 0次 |
码上扫一扫！ |
| 结合特征分布优化的辐射源类增量识别方法 |
| 彭鹏,曹帅,贾勇,姚光乐,王琛,王洪辉,张伟,王祥丰 |
|
|
| (1.成都理工大学 a.计算机与网络安全学院（示范性软件学院）;b.四川省工业互联网智能监测及应用工程技术研究中心; c.机电工程学院，成都 610059;2.电子科技大学 信息与通信工程学院，成都 611730;3.电磁空间安全全国重点实验室，成都 610036;4.华东师范大学 计算机科学与技术学院，上海 200062) |
|
| 摘要: |
| 现有辐射源类增量识别模型受灾难性遗忘的影响，存在识别率低、稳定性-可塑性不平衡等问题。基于分类器增量学习策略，提出了一种结合特征分布优化的辐射源类增量识别方法。通过高斯分布对新类特征进行特征变换，从而改善新类分布，提高新类识别率；利用少量簇内簇间双分布信息，生成与旧类相似的特征分布提高旧类识别率；在高斯分布和双分布生成共同作用下平衡稳定性和可塑性。该方法在手机辐射源数据集和电台辐射源数据集上的平均准确率分别达到了96.38%和94.02%，在平衡稳定性和可塑性的同时有效提升了辐射源识别率。 |
| 关键词: 辐射源识别 类增量学习 灾难性遗忘 特征分布优化 |
| DOI：10.20079/j.issn.1001-893x.250113005 |
|
| 基金项目:国家自然科学基金资助项目(U20B2070)；四川省重点研发项目(2022YFS0531) |
|
| Incremental Identification of Radiation Source Classes Combined with Feature Distribution Optimization |
| PENG Peng,CAO Shuai,JIA Yong,YAO Guangle,WANG Chen,WANG Honghui,ZHANG Wei,WANG Xiangfeng |
| (1a.School of Computer and Network Security(Model Software School);1b.Sichuan Engineering Technology Research Center of Industrial Internet Intelligent Monitoring and Application;1c.School of Electromechanical Engineering，Chengdu University of Technology，Chengdu 610059，China;2.School of Information and Communication Engineering，University of Electronic Science and Technology of China，Chengdu 611730，China;3.National Key Laboratory of Electromagnetic Space Security，Chengdu 610036，China;4.School of Computer Science and Technology，East China Normal University，Shanghai 200062，China) |
| Abstract: |
| The existing models for class incremental recognition of radiation source are affected by catastrophic forgetting，resulting in low recognition rates and imbalance between stability and plasticity.A radiation source class incremental recognition method based on classifier incremental learning and feature distribution optimization is proposed.The new class features are transformed by Gaussian distribution to improve the distribution and increase the recognition rate of the new classes.For the increasing of the recognition rate of the old classes，feature distribution similar to the old classes is generated by utilizing a small amount of inter-and intra-cluster distribution information.The stability and plasticity are eventually balanced through Gaussian distribution and the inter- and intra-cluster distribution.The average accuracy of the proposed method on a mobile phone radiation source dataset and a radio radiation source dataset reaches 96.38% and 94.02%，respectively，which indicates that this method can effectively improve the recognition rate of radiation source while balancing its stability and plasticity. |
| Key words: radiation source identification class incremental learning catastrophic forgetting feature distribution optimization |
《电讯技术》编辑部 