改进DenseNet在电台建链行为识别时的可视化研究

doi:10.12305/j.issn.1001-506X.2021.05.26

系统工程与电子技术 ›› 2021, Vol. 43 ›› Issue (5): 1371-1381.doi: 10.12305/j.issn.1001-506X.2021.05.26

改进DenseNet在电台建链行为识别时的可视化研究

吴子龙(), 陈红(), 雷迎科*()

国防科技大学电子对抗学院, 安徽合肥 230037

收稿日期:2020-05-27 出版日期:2021-05-01 发布日期:2021-04-27
通讯作者: 雷迎科 E-mail:wuzilong@nudt.edu.cn;2392263276@qq.com;22920142204021@stu.xmu.edu.cn
作者简介:吴子龙(1998—), 男, 硕士研究生, 主要研究方向为机器学习、通信信号处理。E-mail: wuzilong@nudt.edu.cn|陈红(1965—), 女, 教授, 硕士研究生导师, 博士, 主要研究方向为现代通信系统、通信信号干扰。E-mail: 2392263276@qq.com|雷迎科(1975—), 男, 副教授, 硕士研究生导师, 博士, 主要研究方向为机器学习、通信信号处理。E-mail: 22920142204021@stu.xmu.edu.cn
基金资助:
国防科技重点实验室基金(6142106180402)

Visualization research on improved DenseNet applied to recognize a radio station's link establishment behavior

Zilong WU(), Hong CHEN(), Yingke LEI*()

School of Electronic Countermeasures, National University of Defense Technology, Hefei 230037, China

Received:2020-05-27 Online:2021-05-01 Published:2021-04-27
Contact: Yingke LEI E-mail:wuzilong@nudt.edu.cn;2392263276@qq.com;22920142204021@stu.xmu.edu.cn

摘要/Abstract

摘要：

在未知通信协议标准条件下, 改进DenseNet可以识别短波通信电台不同的自动建链行为, 这对侦察非协作方电台的通联意图有着重要作用。但是网络模型识别的可靠性需要做进一步的探究。因此, 本文通过网络内部结构的可视化来推断网络模型是否可以有效地提取建链行为信号的深层次特征。由于卷积层的特征维度较大, 不利于观察建链行为信号的深层次特征, 所以对网络内部全连接层进行了可视化分析。实验结果表明, 网络模型可以有效提取同一类别行为信号深层次的相似特征以及不同类别行为信号深层次的差异特征, 并通过热力图进一步提高了实验结果的可信度。

关键词: 可视化, 行为识别, 热力图, DenseNet, 自动建链行为, 短波电台

Abstract:

Under the condition that the communication protocol standard is unknown, the improved DenseNet can recognize a short-wave radio station's different automatic link establishment behaviors, which plays an important role in detecting non-cooperative stations' communication intentions. However, the reliability of network model's recognition needs to be further explored. Therefore, by visualizing the network's internal structure, it is deduced whether the network model can effectively extract the deep features of the link establishment behaviors signals. The feature dimension of the convolutional layer is large, it is not conducive to observe the deep features of the link establishment behavior signal. So the visualization analysis of the full connected layer inside the network is carried out. The experimental results show that the network model can effectively extract the deep features' similarities of the same behavior signals and the deep features' differences of different behavior signals. And experiment of heat maps further improves the credibility of our experimental results.

Key words: visualization, behavior recognition, heat map, DenseNet, automatic link establishment behavior, short-wave radio station

中图分类号:

TP181

吴子龙, 陈红, 雷迎科. 改进DenseNet在电台建链行为识别时的可视化研究[J]. 系统工程与电子技术, 2021, 43(5): 1371-1381.

Zilong WU, Hong CHEN, Yingke LEI. Visualization research on improved DenseNet applied to recognize a radio station's link establishment behavior[J]. Systems Engineering and Electronics, 2021, 43(5): 1371-1381.

图/表 14

表1

表2

图1

表3

表4

表5

图2

图3

图4

图5

图6

图7

图8

图9

参考文献 21

1	HOHMAN F , PARK H , ROBINSON C , et al. SUMMIT: scaling deep learning interpretability by visualizing activation and attribution summarizations[J]. IEEE Trans.on Visualization & Computer Graphics, 2019, 26 (1): 1096- 1106.
2	WU C , GALES M , RAGNI A , et al. Improving interpretability and regularization in deep learning[J]. IEEE/ACM Trans.on Audio, Speech, and Language Processing, 2018, 26 (2): 256- 265. doi: 10.1109/TASLP.2017.2774919
3	ZHANG K , GUO Y R , WANG X S , et al. Multiple feature re-weight DenseNet for image classification[J]. IEEE Access, 2019, 7, 9872- 9880. doi: 10.1109/ACCESS.2018.2890127
4	SONG J M , KIM W , PARK K R . Finger-vein recognition based on deep DenseNet using composite image[J]. IEEE Access, 2019, 7, 66845- 66863. doi: 10.1109/ACCESS.2019.2918503
5	United States Department of Defense. Interoperability and performance standards for medium and high frequency radio systems[P]. U.S. : Standard MIL-STD-188-141B, 1999.
6	游翔, 葛卫丽. 飞信协议识别与多元通联关系提取方法[J]. 现代电子技术, 2014, 37 (21): 19- 23.
	YOU X , GE W L . Protocol identification and multi-conversation relationship extraction in Fetion[J]. Modern Electronics Technique, 2014, 37 (21): 19- 23.
7	PROTAS É , BRATTI J D , GRYA J F O , et al. Visualization methods for image transformation convolutional neural networks[J]. IEEE Trans.on Neural Networks and Learning Systems, 2019, 30 (7): 2231- 2243. doi: 10.1109/TNNLS.2018.2881194
8	DUAN P H, KANG X D, LI S T. Convolutional neural network for natural color visualization of hyperspectral images[C]//Proc. of the IEEE International Geoscience and Remote Sensing Symposium, 2019: 3372-3375.
9	ZEILER, MATTHEW D, ROB F. Visualizing and understanding convolutional networks[C]//Proc. of the European Conference on Computer Vision, 2014: 818-833.
10	SPRINGENBERG J T, DOSOVITSKIY A, BROX T, et al. Striving for simplicity: the all convolutional net[C]//Proc. of the International Conference on Learning Representations, 2014.
11	SELVARAJU R R , COGSWELL M , DAS A , et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[J]. International Journal of Computer Vision, 2020, 128 (2): 336- 359. doi: 10.1007/s11263-019-01228-7
12	ZHANG Q S, WU Y N, ZHU S C. Interpretable convolutional neural networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8827-8836.
13	WU T F, SUN W, LI X L, et al. Towards interpretable R-CNN by unfolding latent structures[EB/OL]. [2020-05-01]. https://arxiv.org/abs/1711.05226.
14	ZHANG Q S , ZHU S C . Visual interpretability for deep learning: a survey[J]. Journal of Zhejiang University Science C, 2018, 19 (1): 27- 39. doi: 10.1631/FITEE.1700808
15	RIBEIRO M T, SINGH S, GUESTRIN C. "Why should I trust you?": explaining the predictions of any classifier[C]//Proc. of the North American Chapter of the Association for Computational Linguistics, 2016: 97-101.
16	NICOLAS P, MCDANIEL P D. Deep k-nearest neighbors: towards confident, interpretable and robust deep learning[EB/OL]. [2020-05-01]. https://arxiv.org/abs/1803.04765.
17	RU C, CAO J W. Radar emitter identification with bispectrum based LBP and extreme learning machine[C]//Proc. of the IEEE International Conference on Digital Signal Processing, 2018.
18	WANG G B , GU F S , REHAB I , et al. A sparse modulation signal bispectrum analysis method for rolling element bearing diagnosis[J]. Mathematical Problems in Engineering, 2018, 4, 1- 12.
19	MENDIS G J , JIN W K , MADANAYAKE A . Deep learning based radio-signal identification with hardware design[J]. IEEE Trans.on Aerospace and Electronic Systems, 2019, 55 (5): 2516- 2531. doi: 10.1109/TAES.2019.2891155
20	李雨珊, 谢非佚, 陈松林, 等. 适合终端的射频指纹信号特征提取及识别[J]. 通信技术, 2018, 51 (1): 63- 66.
	LI Y S , XIE F Y , CHEN S L , et al. Feature extraction and recognition of radio frequency fingerprint signal suitable for terminal[J]. Communications Technology, 2018, 51 (1): 63- 66.
21	O'SHEA T J , ROY T , CLANCY T C . Over-the-air deep learning based radio signal classification[J]. IEEE Journal of Selected Topics in Signal Processing, 2018, 12 (1): 168- 179. doi: 10.1109/JSTSP.2018.2797022

建链行为种类	功能
呼叫行为	一方电台想要和另一方电台建立通信(另一方电台可以是一个、一组或其余所有电台)
握手行为	当接收到另一方电台呼叫时, 电台想要继续通信而进行的行为
通知行为	一方电台以通知的形式向所有其他电台发出信息
时间校准行为	电台进行时间校准的通联行为
组时间广播行为	一方电台向处于同一个通信网络组的其余电台进行时间广播
广播行为	一方电台指示另一方电台执行某一命令
扫描呼叫行为	已建立过通信链路的双方电台再次进行通信所进行的呼叫行为

层	输入矩阵维度大小	输出矩阵维度大小
输入层	(5 888, 32)	(5 888, 32)
BN+ReLu+Conv	(5 888, 32)	(5 888, 32)
特征复用层	(5 888, 32), (5 888, 32)	(5 888, 64)
BN+ReLu+Conv	(5 888, 64)	(5 888, 32)
特征复用层	[(5 888, 32)]×3	(5 888, 96)
BN+ReLu+Conv	(5 888, 96)	(5 888, 32)
特征复用层	[(5 888, 32)]×4	(5 888, 128)
BN+ReLu+Conv	(5 888, 128)	(5 888, 32)
输出层	[(5 888, 32)]×5	(5 888, 160)

训练样本数目	识别准确率/%
1 400	56.34
2 800	63.11
4 200	65.88
5 600	68.32
7 000	69.83
8 400	70.79
9 800	71.23
11 200	72.94
12 600	73.38

训练样本数目	识别准确率/%
1 400	86.46
2 800	92.30
4 200	94.16
5 600	96.44
7 000	97.48
8 400	98.19
9 800	98.26
11 200	98.38
12 600	99.14

训练样本数目	识别准确率/%
1 400	99.58
2 800	99.83
4 200	99.85
5 600	99.89
7 000	99.90
8 400	99.93
9 800	99.98
11 200	99.98
12 600	99.99

改进DenseNet在电台建链行为识别时的可视化研究

Visualization research on improved DenseNet applied to recognize a radio station's link establishment behavior

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 21

相关文章 9

编辑推荐

Metrics

本文评价

[1]	董浩洋, 张东戈, 齐宁, 禹明刚, 牛彦杰. 基于关注势的战场态势热力图构建方法[J]. 系统工程与电子技术, 2021, 43(1): 121-129.
[2]	蒲玮, 李雄. 基于EA流程图的指挥流程信息可视化方法[J]. 系统工程与电子技术, 2017, 39(1): 215-222.
[3]	张才坤, 朱战霞, 冯琦, 张堃. 基于区间数雷达图的可视化空战威胁评估[J]. 系统工程与电子技术, 2016, 38(5): 1052-.
[4]	胡华全，吴玲达，杨超，于荣欢，宋汉辰. 卫星时变拓扑网络多视图可视化分析框架[J]. 系统工程与电子技术, 2014, 36(2): 312-316.
[5]	胡华全，吴玲达，杨超，宋汉辰. 基于图序列的卫星网络时变拓扑可视分析[J]. 系统工程与电子技术, 2014, 36(10): 2065-2070.
[6]	毕晓君，李博. 基于决策偏好信息的n维Pareto前沿可视化[J]. Journal of Systems Engineering and Electronics, 2013, 35(3): 557-563.
[7]	于荣欢, 吴玲达, 瞿师. 组网雷达探测能力的并行计算与可视化方法研究[J]. Journal of Systems Engineering and Electronics, 2011, 33(11): 2512-2516.
[8]	秦大国1,2，陈小武1，李波1，刘立娜2. 空间态势计算与可视化建模[J]. Journal of Systems Engineering and Electronics, 2009, 31(12): 2904-2908.
[9]	杨超,徐江斌,赵健,吴玲达. 基于多层等值面的电磁环境三维可视化研究[J]. Journal of Systems Engineering and Electronics, 2009, 31(11): 2767-2772 .