Open-MUSIC：基于度量学习与特征子空间投影的电磁目标开集识别算法

doi:10.12263/DZXB.20210829

电子学报 ›› 2022, Vol. 50 ›› Issue (6): 1310-1318.DOI: 10.12263/DZXB.20210829

所属专题：电磁频谱智能+

Open-MUSIC：基于度量学习与特征子空间投影的电磁目标开集识别算法

杨柳, 利强, 邵怀宗

电子科技大学信息与通信工程学院，四川成都 611731

收稿日期:2021-07-01 修回日期:2022-01-02 出版日期:2022-06-25
- 通讯作者:
- 利强
- 作者简介:
- 杨　柳　男，1999年出生，安徽芜湖人.2020年在电子科技大学获得学士学位，现为硕博连读生.主要研究方向为知识引导的机器学习.E-mail: 1070032777@qq.com
  利　强（通讯作者）　男，1982年出生，四川成都人.副教授、博士生导师.分别于2005年、2008年在电子科技大学获得学士和硕士学位，于2012年在香港中文大学获得博士学位.主要研究方向为无线通信优化算法、电磁频谱智能感知等. E-mail:lq@uestc.edu.cn
  邵怀宗　男，1967年出生，四川巴中人.教授、博士生导师.1992年在长春理工大学获得学士学位，1998年在四川大学获得硕士学位，2003年在电子科技大学获得博士学位.主要研究方向为无线通信、电子对抗、新体制通信、人工智能等.E-mail: hzshao@uestc.edu.cn
- 基金资助:
- 国家自然科学基金 (62171110)

Open-MUSIC: An Open Set Recognition Algorithm of Electromagnetic Target Based on Metric Learning and Feature Subspace Projection

YANG Liu, LI Qiang, SHAO Huai-zong

School of Information and Communication Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan 611731, China

Received:2021-07-01 Revised:2022-01-02 Online:2022-06-25 Published:2022-06-25
- Corresponding author:
- LI Qiang
- Supported by:
- National Natural Science Foundation of China (62171110)

摘要/Abstract

摘要：

在越来越复杂的电磁频谱环境中，要想实现对频谱资源的管控，首先要判断发送信号的辐射源是否是己方已知的.针对此问题，本文提出了一种基于计算特征子空间投影比值的算法Open-MUSIC（MUltiple SIgnal Classification），通过神经网络获得已知类特征表示；进而得到已知类特征矩阵的两个正交子空间；以特征在两个子空间内的投影比值为指标，对辐射源信号样本是否为已知做判决.在3个数据集上的仿真表明，Open-MUSIC算法的性能在电磁数据集上较其他方法提升了3%以上.

关键词: 开集识别, 特征子空间分解, 中心损失, 投影

Abstract:

In the increasingly complex electromagnetic spectrum environment, in order to realize the management and control of spectrum resources, it is necessary to determine whether the received signal is from the known or unknown radiation source. To tackle this problem, this paper proposes an algorithm named Open-MUSIC(MUltiple SIgnal Classification) to discriminate the known and unknown sources. The key idea of Open-MUSIC is to form the feature space from the known classes via a judiciously designed neural network, and then the feature space is decomposed into two orthogonal subspaces, namely the range subspace and the null subspace. Based on the projection ratio of the test signal's feature onto the two subspaces, we can accurately discriminate the known and the unknown radiation sources. Experiments on three datasets show that the performance of the Open-MUSIC is improved by more than 3% on electromagnetic data sets compared to other methods.

Key words: open set recognition, feature subspace decomposition, center loss, projection

中图分类号:

TN911.7

杨柳, 利强, 邵怀宗. Open-MUSIC：基于度量学习与特征子空间投影的电磁目标开集识别算法[J]. 电子学报, 2022, 50(6): 1310-1318.

Liu YANG, Qiang LI, Huai-zong SHAO . Open-MUSIC: An Open Set Recognition Algorithm of Electromagnetic Target Based on Metric Learning and Feature Subspace Projection[J]. Acta Electronica Sinica, 2022, 50(6): 1310-1318.

图/表 12

参考文献 24

1	BENDALEA, BOULTT E. Towards open set deep networks[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 1563-1572.
2	BENDALEA, BOULTT. Towards open world recognition[C]//2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 1893-1902.
3	KRIZHEVSKYA, SUTSKEVERI, HINTONG E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
4	O'SHEAT J, ROYT, CLANCYT C. Over-the-air deep learning based radio signal classification[J]. IEEE Journal of Selected Topics in Signal Processing, 2018, 12(1): 168-179.
5	HEK M, ZHANGX Y, RENS Q, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
6	PENGS L, JIANGH Y, WANGH X, et al. Modulation classification based on signal constellation diagrams and deep learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 30(3): 718-727.
7	DUANS R, CHENK, YUX, et al. Automatic multicarrier waveform classification via PCA and convolutional neural networks[J]. IEEE Access, 2018, 6: 51365-51373.
8	WONGL J, HEADLEYW C, MICHAELSA J. Specific emitter identification using convolutional neural network-based IQ imbalance estimators[J]. IEEE Access, 2019, 7: 33544-33555.
9	HUANGS, CHAIL, LIZ N, et al. Automatic modulation classification using compressive convolutional neural network[J]. IEEE Access, 2019, 7: 79636-79643.
10	ZHANGX W, SEYFIT, JUS T, et al. Deep learning for interference identification: Band, training SNR, and sample selection[C]//2019 IEEE 20th International Workshop on Signal Processing Advances in Wireless Communications. Cannes: IEEE, 2019: 1-5.
11	LIANGF, SHENC, YUW, et al. Towards optimal power control via ensembling deep neural networks[J]. IEEE Transactions on Communications, 2020, 68(3): 1760-1776.
12	CEVIKALPH, TRIGGSB, FRANCV. Face and landmark detection by using cascade of classifiers[C]//2013 10th IEEE International Conference and Workshops on Automatic Face and Gesture Recognition. Shanghai: IEEE, 2013: 1-7.
13	CEVIKALPH. Best fitting hyperplanes for classification[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1076-1088.
14	DHAMIJAA R, GÜNTHERM, BOULTT E. Reducing Network Agnostophobia[EB/OL]. (2018)[2021]. .
15	SHUL, XUH, LIUB. DOC: Deep Open Classification of Text Documents[EB/OL]. (2017)[2021]. .
16	LIUZ W, MIAOZ Q, ZHANX H, et al. Large-scale long-tailed recognition in an open world[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Long Beach: IEEE, 2019: 2532-2541.
17	HASSENM, CHANP K. Learning a neural-network-based representation for open set recognition[C]//Proceedings of the 2020 SIAM International Conference on Data Mining. Philadelphia, PA: Society for Industrial and Applied Mathematics, 2020: 154-162.
18	GEZ Y, DEMYANOVS, GARNAVIR. Generative OpenMax for multi-class open set classification[C]//Proceedings of the British Machine Vision Conference 2017. London: British Machine Vision Association, 2017: 42.1-42.12.
19	YUY, QUW Y, LIN, et al. Open category classification by adversarial sample generation[C]//Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence. California: International Joint Conferences on Artificial Intelligence Organization, 2017: 469.
20	CHENG Y, PENGP X, WANGX Q, et al. Adversarial reciprocal points learning for open set recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, DOI:10.1109/TPAMI.2021.3106743.
21	DE MAESSCHALCKR, JOUAN-RIMBAUDD, MASSARTD L. The mahalanobis distance[J]. Chemometrics and Intelligent Laboratory Systems, 2000, 50(1): 1-18.
22	WANGX, HANX T, HUANGW L, et al. Multi-similarity loss with general pair weighting for deep metric learning[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). Long Beach: IEEE, 2019: 5017-5025.
23	WENY D, ZHANGK P, LIZ F, et al. A discriminative feature learning approach for deep face recognition[C]//European Conference On Computer Vision. Cham: Springer, 2016: 499-515..
24	MENSINKT, VERBEEKJ, PERRONNINF, et al. Distance-based image classification: Generalizing to new classes at near-zero cost[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(11): 2624-2637.

层名称	卷积核大小/神经元个数	卷积核深度	激励函数
卷积层	5*5	32	PReLU
卷积层	5*5	32	PReLU
池化层	2*2	—	—
卷积层	5*5	64	PReLU
卷积层	5*5	64	PReLU
池化层	2*2	—	—
卷积层	5*5	128	PReLU
卷积层	5*5	128	PReLU
池化层	2*2	—	—
全连接层	32	—	PReLU
全连接层	6	—	—

层名称	卷积核大小/神经元个数	卷积核深度	激励函数
卷积层	5*5	32	PReLU
卷积层	5*5	32	PReLU
池化层	2*2	—	—
卷积层	5*5	64	PReLU
卷积层	5*5	64	PReLU
池化层	2*2	—	—
卷积层	5*5	128	PReLU
卷积层	5*5	128	PReLU
池化层	2*2	—	—
全连接层	32	—	PReLU
全连接层	6	—	—

层名称	卷积核大小/神经元个数	卷积核深度	激励函数
卷积层	3*3	32	ReLU
池化层	2*2	—	—
卷积层	3*3	64	ReLU
池化层	2*2	—	—
卷积层	5*5	128	ReLU
池化层	2*2	—	—
全连接层	512	—	ReLU
全连接层	256	—	ReLU
全连接层	128	—	ReLU
全连接层	55	—	—

层名称	卷积核大小/神经元个数	卷积核深度	激励函数
卷积层	3*3	32	ReLU
池化层	2*2	—	—
卷积层	3*3	64	ReLU
池化层	2*2	—	—
卷积层	5*5	128	ReLU
池化层	2*2	—	—
全连接层	512	—	ReLU
全连接层	256	—	ReLU
全连接层	128	—	ReLU
全连接层	55	—	—

层名称	卷积核大小/神经元个数	卷积核深度	激励函数
卷积层	3*3	32	ReLU
卷积层	3*3	32	ReLU
池化层	2*2	—	—
卷积层	3*3	64	ReLU
卷积层	3*3	64	ReLU
池化层	2*2	—	—
卷积层	3*3	128	ReLU
卷积层	3*3	128	ReLU
池化层	2*2	—	—
卷积层	3*3	256	ReLU
卷积层	3*3	256	ReLU
池化层	2*2	—	—
全连接层	512	—	ReLU
全连接层	256	—	ReLU
全连接层	128	—	ReLU
全连接层	5	—	—

Open-MUSIC：基于度量学习与特征子空间投影的电磁目标开集识别算法

Open-MUSIC: An Open Set Recognition Algorithm of Electromagnetic Target Based on Metric Learning and Feature Subspace Projection

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 24

相关文章 15

编辑推荐

Metrics

算法	评估指标
算法	F-measure	Accuracy	AUC-ROC
OpenMax	0.791	0.800	0.981
OLTR	0.653	0.703	0.986
ii-loss	0.861	0.842	0.983
ARPL	0.966	0.966	0.996
Open-MUSIC	0.937	0.938	0.984

[1]	刘少鹏, 赵慧民, 洪佳明, 吴晓航, 许发宝, 欧阳佳, 梁鹏, 熊建斌. 面向医学图像生成的鲁棒条件生成对抗网络[J]. 电子学报, 2023, 51(2): 427-437.
[2]	刘杰, 游品鸿, 田明, 刘金凤. 基于局部投影的视差图像拼接平滑优化[J]. 电子学报, 2022, 50(6): 1451-1456.
[3]	刘秋红, 许漫坤, 李天昀. 基于ML的多径信道下直扩信号PN码和信道的联合盲估计[J]. 电子学报, 2021, 49(8): 1480-1488.
[4]	邓翔宇, 杨雅涵. 纸质简谱图像的分割、倾斜校正及音符歌词提取[J]. 电子学报, 2021, 49(4): 716-728.
[5]	李燕, 严永财. 基于同心圆阵列的结构光主动视觉系统标定算法[J]. 电子学报, 2021, 49(3): 536-541.
[6]	李东瑾, 杨瑞娟, 李晓柏, 董睿杰. 基于核协同表示与鉴别投影的辐射源调制识别[J]. 电子学报, 2020, 48(9): 1695-1702.
[7]	张彩霞, 王子涵, 文成林, 刘国文, 余伟. 样本空间基于多级高维特征表示的微小故障诊断[J]. 电子学报, 2020, 48(8): 1647-1654.
[8]	蔺想红, 郑鉴洋, 王向文, 马慧芳. 基于深度学习网络的神经元自适应投影分类方法[J]. 电子学报, 2020, 48(7): 1321-1329.
[9]	赖锦祥, 雷亮, 阮杨, 邓凯元, 徐初东, 周金运. 基于等积环投影与Zernike矩的快速模板匹配[J]. 电子学报, 2019, 47(6): 1324-1330.
[10]	李玉, 王亚琼, 赵雪梅, 赵泉华. 基于随机投影深度函数的停车场车辆提取方法[J]. 电子学报, 2019, 47(2): 322-330.
[11]	张素玲, 陈胜垚, 席峰, 刘中. 亚奈奎斯特采样雷达的运动目标回波信号的快速重构[J]. 电子学报, 2019, 47(10): 2098-2107.
[12]	陈泽亮, 彭华, 巩克现, 于沛东, 王伟年. 误码条件下LDPC码参数的盲估计[J]. 电子学报, 2018, 46(3): 652-658.
[13]	彭亚丽, 刘侍刚, 孙增国, 洪灵, 曹菡. 迭代二次规划遮挡点恢复[J]. 电子学报, 2018, 46(11): 2733-2737.
[14]	罗婷, 李孟飞, 赵云松. 基于二次有理分式拟合多能投影曲线的X射线能谱估计方法[J]. 电子学报, 2018, 46(11): 2580-2587.
[15]	唐波, 张玉, 张浩. 基于交替投影的MIMO雷达信号盲分离算法研究[J]. 电子学报, 2017, 45(9): 2092-2097.

算法	评估指标
算法	F-measure	Accuracy	AUC-ROC
OpenMax	0.415	0.470	0.577
OLTR	0.727	0.639	0.833
ii-loss	0.911	0.906	0.935
ARPL	0.713	0.621	0.747
Open-MUSIC	0.941	0.937	0.969

算法	评估指标
算法	F-measure	Accuracy	AUC-ROC
OpenMax	0.415	0.470	0.577
OLTR	0.727	0.639	0.833
ii-loss	0.911	0.906	0.935
ARPL	0.713	0.621	0.747
Open-MUSIC	0.941	0.937	0.969

算法	评估指标
算法	F-measure	Accuracy	AUCROC
OpenMax	0.700	0.623	0.915
OLTR	0.653	0.703	0.881
ii-loss	0.813	0.815	0.907
ARPL	0.680	0.626	0.803
Open-MUSIC	0.916	0.911	0.956

算法	评估指标
算法	F-measure	Accuracy	AUCROC
OpenMax	0.700	0.623	0.915
OLTR	0.653	0.703	0.881
ii-loss	0.813	0.815	0.907
ARPL	0.680	0.626	0.803
Open-MUSIC	0.916	0.911	0.956