低干噪比下多特征融合的通信干扰识别方法

doi:10.12305/j.issn.1001-506X.2026.03.30

摘要/Abstract

摘要：

针对低干噪比下通信干扰信号特征微弱、单一特征表征能力不足导致识别准确率低的问题，提出一种基于YOLOv8和多特征融合网络的通信干扰信号识别方法。首先，提出一种基于投影变换的特征图像提取方法，根据不同特征对于不同信号的不同敏感度，选取具有强互补特性的双谱变换、希尔伯特黄变换、短时傅里叶变换作为识别特征，并通过投影变换将其统一为适用于二维卷积网络的特征图；其次，利用YOLOv8网络强大的特征学习能力，同时从上述3种异构特征图中并行提取深层判别信息；然后，设计了一种决策级全连接融合网络，对YOLOv8输出的概率向量进行加权整合，实现最终分类决策。实验结果表明，该方法显著提升了低干噪比下的识别性能，尤其在?20 dB的低干噪比恶劣条件下，识别准确率达到48.18%，相较于传统方法以及单一特征方法分别提高了40.60%和31.86%。

关键词: 干扰识别, 多特征融合, 深度学习, YOLOv8

Abstract:

To address the issue of low recognition accuracy for communication interference signals under low interference-to-noise ratio conditions, which is caused by weak signal features and the insufficient representation capability of single feature, a communication interference signal recognition method based on YOLOv8 and multi-feature fusion networks is proposed. First, a feature image extraction method based on projection transformation is introduced. Based on the varying sensitivities of different features to different signals, bispectrum transform, Hilbert-Huang transform, and short-time Fourier transform, which possess strong complementary characteristics, are selected as recognition features, and they are unified into feature maps suitable for two-dimensional convolutional neural networks via projection transformation. Secondly, leveraging the powerful feature learning capability of the YOLOv8 network, deep discriminative information is extracted in parallel from the aforementioned three heterogeneous feature maps. Then, a decision-level fully connected fusion network is designed to perform weighted integration of the probability vectors output by YOLOv8, thereby achieving the final classification decision. Experimental results demonstrate that this method significantly enhances the recognition performance under low interference-to-noise ratio conditions. Particularly, under the adverse condition of a ?20 dB low interference-to-noise ratio, the recognition accuracy reaches 48.18%, representing improvements of 40.60% and 31.86% over traditional methods and single-feature methods, respectively.

Key words: interference recognition, multi-feature fusion, deep learning, YOLOv8

中图分类号:

TN 92

田弘宇, 葛松虎, 郭宇, 崔中普, 梁潇. 低干噪比下多特征融合的通信干扰识别方法[J]. 系统工程与电子技术, 2026, 48(3): 1061-1071.

Hongyu TIAN, Songhu GE, Yu GUO, Zhongpu CUI, Xiao LIANG. Communication interference recognition method based on multi-feature fusion under low interference-to-noise ratio conditions[J]. Systems Engineering and Electronics, 2026, 48(3): 1061-1071.

图/表 15

表1

干扰信号模型"

干扰类型	数学模型	参数	参数含义
STJ	$x(t) = \sqrt {{A_j}} \exp [{\mathrm{j}}(2\text{π} {f_j}t + {\varphi _j})]$	${A_j}$	发射的干扰信号功率
		${f_j}$	干扰信号中心频率
		${\varphi _j}$	干扰信号初始相位，服从[0,2π）均匀分布
MTJ	$x(t) = \displaystyle\sum\limits_{i = 1}^M {\sqrt {{A_{{j_{\text{i}}}}}} \exp [{\mathrm{j}}(2\text{π} {f_{{j_i}}}t + {\varphi _{{j_i}}})]} $	$M$	单音干扰的数量
		${A_{{j_i}}}$	发射的干扰信号功率
		${f_{{j_i}}}$	干扰信号中心频率
		${\varphi _{{j_i}}}$	干扰信号初始相位，服从[0,2π）均匀分布且互相独立同分布
LFM	$ x(t) = \sqrt {{A_j}} \exp [{\mathrm{j}}(2\text{π} {f_j}t + \text{π} k{t^2} + {\varphi _j}] $	$k$	线性调频的斜率参数
NoiFM	$ x(t) = \sqrt {{A_j}} \exp [{\mathrm{j}}(2\text{π} {f_j}t + 2\text{π} {K_{{\mathrm{FM}}}}\displaystyle\int\nolimits_0^t {u(\tau ){\mathrm{d}}\tau } + {\varphi _j})] $	${K_{{\mathrm{FM}}}}$	调频比例常数
NoiFM		$u(\tau )$	带限高斯白噪声
NoiAM	$x(t) = [{U_0} + {U_n}(t)]\exp [{\mathrm{j}}(2\text{π} {f_j}t + {\varphi _j})]$	${U_0}$	发射的干扰信号功率
PBNJ	$x(t) = {U_n}(t)\exp [{\mathrm{j}}(2\text{π} {f_j}t + {\varphi _j})]$	$ {U_n}(t) $	带限高斯白噪声
BPSK	$x(t) = \sqrt {{A_j}} \exp [{\mathrm{j}}(2\text{π} {f_j}t + {\varphi _b} + {\varphi _j})]$	${\varphi _b}$	BPSK调制相位，取值为${\varphi _b} = 0,\pi $，分别表示比特0、1
QPSK	$x(t) = \sqrt {{A_j}} \exp [{\mathrm{j}}(2\text{π} {f_j}t + {\varphi _q} + {\varphi _j})]$	${\varphi _q}$	QPSK调制相位，取值为${\varphi _q} = 0,\dfrac{\text{π} }{2},\text{π} ,\dfrac{{3\text{π} }}{2}$，分别表示比特00、01、10、11
2ASK	$x(t) = \sqrt {{A_j}} m(t)\exp [{\mathrm{j}}(2\text{π} {f_j}t + {\varphi _j})]$	$m(t)$	$m(t) = 0,1$，分别表示比特0、1
2FSK	$x(t) = \sqrt {{A_j}} {m_j}(t)\exp [{\mathrm{j}}(2\text{π} {f_j}t + {\varphi _j})]$ ${m_j}(t) = \exp [{\mathrm{j}}(m(t) \cdot 2\text{π} {f_0}t + \overline {m(t)} \cdot 2\text{π} {f_1}t)]$	$\overline {m(t)} $	$m(t)$的相反数
2FSK		${m_j}(t)$	调制信号
MSK	$ x(t) = \sqrt {{A_j}} \exp \left[{\mathrm{j}}\left(2\text{π} {f_j}t + \dfrac{{\text{π} {a_k}}}{{2{T_s}}}t + {\varphi _k}\right)\right] $ $ k{T_s} \leqslant t \leqslant (k + 1){T_s},\;k = 0,1 \cdots $	${a_k}$	${a_k} = 1, - 1$，分别表示比特1、0

表1

图1

图2

图3

图4

图5

图6

图7

表2

表3

图8

图9

图10

图11

表4

参考文献 25

1	RAHMAN M J A, KRUNZ M, ERWIN R. Interference mitigation using spectrum sensing and dynamic frequency hopping[C]//Proc. of the IEEE International Conference on Communications , 2012: 4421−4425.
2	XIAO L, LIU J L, LI Q D, et al. User-centric view of jamming games in cognitive radio networks[J]. IEEE Trans. on Information Forensics and Security, 2015, 10 (12): 2578- 2590. doi: 10.1109/TIFS.2015.2467593
3	DABCEVIC K, MUGHAL M O, MARCENARO L, et al. Cognitive radio as the facilitator for advanced communications electronic warfare solutions[J]. Journal of Signal Processing Systems, 2016, 83 (1): 29- 44. doi: 10.1007/s11265-015-1050-0
4	MUGHAL M O, NAWAZ T, MARCENARO L, et al. Cyclostationary-based jammer detection algorithm for wide-band radios using compressed sensing[C]//Proc. of the IEEE Global Conference on Signal and Information Processing, 2015: 280−284.
5	NUNES F D, SOUSA F M G. Jamming detection in GNSS signals using the sample covariance matrix[C]//Proc. of the 6th ESA Workshop on Satellite Navigation Technologies (Navitec 2012) & European Workshop on GNSS Signals and Signal Processing, 2012.
6	邵正途, 冷毅, 葛代河, 等. 基于小波包分解和随机森林特征选择的雷达有源干扰识别[J]. 现代雷达, 2024, 46 (12): 45- 52.
	SHAO Z T, LENG Y, GE D H, et al. Radar active interference recognition based on wavelet packet decomposition and random forest feature selection[J]. Modern Radar, 2024, 46 (12): 45- 52.
7	KONG L, XU Z J, WANG J M, et al. A novel algorithm for jamming recognition in wireless communication[C]//Proc. of the 6th International Congress on Image and Signal Processing, 2013: 1473−1477.
8	HAN H, REN Z Y, LI L, et al. Automatic modulation classification based on deep feature fusion for high noise level and large dynamic input[J]. Sensors, 2021, 21, 2117. doi: 10.3390/s21062117
9	SWINNEY C J, WOODS J C. GNSS jamming classification via CNN, transfer learning & the novel concatenation of signal representations[C]//Proc. of the International Conference on Cyber Situational Awareness, Data Analytics and Assessment, 2021.
10	WANG G S, REN Q H, SU Y Z. The interference classification and recognition based on SF-SVM algorithm[C]//Proc. of the IEEE 9th International Conference on Communication Software and Networks, 2017: 835−841.
11	ARJOUNE Y, SALAHDINE F, ISLAM Md S, et al. A novel jamming attacks detection approach based on machine learning for wireless communication[C]//Proc. of the International Conference on Information Networking, 2020: 459−464.
12	O’SHEA T J, CORGAN J, CLANCY T C. Convolutional radio modulation recognition networks[M]//JAYNE C, ILIADIS L, ed. Engineering Applications of Neural Networks. Cham: Springer International Publishing, 2016: 213−226.
13	LIU Q, ZHANG W. Deep learning and recognition of radar jamming based on CNN[C]//Proc. of the 12th International Symposium on Computational Intelligence and Design, 2019: 208−212.
14	魏迪, 曾海彬, 洪锋, 等. 基于LSTM网络和特征融合的通信干扰识别[J]. 电讯技术, 2022, 62 (4): 450- 456. doi: 10.3969/j.issn.1001-893x.2022.04.007
	WEI D, ZENG H B, HONG F, et al. Communication interference recognition based on LSTM network and feature fusion[J]. Telecommunication Technology, 2022, 62 (4): 450- 456. doi: 10.3969/j.issn.1001-893x.2022.04.007
15	STEINER J, PESIK J. Machine learning image recognition for GNSS jamming signals categorization[J]. Neural Network World, 2024, 34 (6): 341- 360. doi: 10.14311/NNW.2024.34.019
16	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770−778.
17	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 779−788.
18	ANNA T Q, REESE R D. YOLO for radio frequency signal classification[C]//Proc. of the IEEE International Symposium on Dynamic Spectrum Access Networks, 2024: 177−178.
19	LI R D, HU J H, LI S Q, et al. Blind detection of communication signals based on improved YOLO3[C]//Proc. of the 6th International Conference on Intelligent Computing and Signal Processing, 2021: 424−429.
20	YAN W D, LIU W, BI H B, et al. YOLO-PD: abnormal signal detection in gas pipelines based on improved YOLOv7[J]. IEEE Sensors Journal, 2023, 23 (17): 19737- 19746. doi: 10.1109/JSEN.2023.3296131
21	WANG C Y, BOCHKOVSKIY A, LIAO H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2023: 7464−7475.
22	SARKAR S, GUO D N, CABRIC D. RadYOLOLet: radar detection and parameter estimation using YOLO and wavelet[J]. IEEE Trans. on Cognitive Communications and Networking, 2025, 11 (4): 2340- 2355.
23	PENG R H, WU X R, WANG G H, et al. Intelligent recognition and information extraction of radar complex jamming based on time-frequency features[J]. Journal of Systems Engineering and Electronics, 2024, 35 (5): 1148- 1166. doi: 10.23919/JSEE.2024.000073
24	JOCHER G, CHAURASIA A, QIU J. Ultralytics YOLO[EB/OL]. [2024-08-31]. https://github.com/ultralytics/ultralytics.
25	王鹏宇. 基于深度学习的无线通信干扰识别技术研究[D]. 成都: 电子科技大学, 2023.
	WANG P Y. Research on wireless communication interference recognition technology based on deep learning[D]. Chengdu: University of Electronic Science and Technology of China, 2023.

参数	参数值
采样频率/MHz	10
INR范围/dB	−20~10
INR步长/dB	2
信号采样点数	131 328

干扰类别	参数	参数值	干扰类别	参数	参数值
STJ	干扰频率/MHz	2	BPSK	码元速率/（kbit/s）	500
MTJ	干扰频率/MHz	1.5 2 3	QPSK	码元速率/（kbit/s）	500
LFM	调频系数/（MHz/s）	500	2ASK	码元速率/（kbit/s）	500
LFM	调频带宽/MHz	1	2ASK	码元速率/（kbit/s）	500
NoiFM	调频系数/（Hz/v）	500	2FSK	码元速率/（kbit/s）	500
NoiFM	调频带宽/MHz	1	2FSK	码元速率/（kbit/s）	500
NoiAM	调频带宽/MHz	1	MSK	码元速率/（kbit/s）	500
PBNJ	干扰带宽/MHz	1

方法	INR/dB
方法	10	0	−10	−20
BT+HHT+STFT	100	99.55	88.18	48.18
BT+HHT	94.55	87.73	53.64	22.27
BT+STFT	100	99.55	85.00	32.27
HHT+STFT	100	99.55	77.73	43.18
BT	84.55	84.09	38.64	10.45
HHT	80.45	29.55	15.45	6.36
STFT	100	98.64	69.09	16.82

[1]	程鲲, 雷洪涛, 吕志轩. 改进YOLOv8的轻量级无人机跟踪方法[J]. 系统工程与电子技术, 2026, 48(3): 737-750.
[2]	张龙, 朱连宏, 杨波, 雷震, 冯轩铭. MFA-Net：一种面向复杂对抗环境的反舰导弹智能识别多模态自适应融合网络[J]. 系统工程与电子技术, 2026, 48(3): 894-907.
[3]	包玉刚, 贾皓翔, 赵旦峰. 基于高阶递归网络的单幅图像去雨滴模型[J]. 系统工程与电子技术, 2026, 48(1): 12-21.
[4]	陈海青, 汪刘应, 刘顾, 王龙, 葛超群, 陈孟州. 合成孔径雷达图像去噪算法研究进展[J]. 系统工程与电子技术, 2026, 48(1): 94-105.
[5]	李开明, 张袁鹏, 罗迎, 代肖楠. 弹道导弹雷达目标识别研究进展[J]. 系统工程与电子技术, 2025, 47(9): 2870-2889.
[6]	季策, 马相宇, 牟晓宇, 赵家毅. TS-GRU-VTA：基于深度学习的车辆信道估计方案[J]. 系统工程与电子技术, 2025, 47(9): 3093-3098.
[7]	闫小伟, 凌冲, 石胜斌. 地表未爆子弹药快速检测系统设计与实现[J]. 系统工程与电子技术, 2025, 47(8): 2639-2645.
[8]	张琬滢, 高由兵, 李泽一, 李鹏飞, 张伟. 基于双路径特征融合网络的背景信号抑制算法[J]. 系统工程与电子技术, 2025, 47(7): 2406-2413.
[9]	张新征, 闫梦可, 朱晓林. 噪声伪标签容忍的半监督SAR目标识别[J]. 系统工程与电子技术, 2025, 47(6): 1796-1805.
[10]	姜智杰, 宋恒, 胡楠, 段兰茜, 曹平. 隧道环境毫米波雷达目标识别与分类算法[J]. 系统工程与电子技术, 2025, 47(5): 1453-1460.
[11]	聂千祁, 沙明辉, 朱应申, 王崇宇, 崔念强. 基于盲源分离结合奇异谱分析的雷达多分量信号识别方法[J]. 系统工程与电子技术, 2025, 47(4): 1168-1175.
[12]	袁磊, 杨艳娟, 郭毅, 戴鹏. 相关噪声下基于深度学习的LDPC码码率半盲识别算法[J]. 系统工程与电子技术, 2025, 47(4): 1335-1345.
[13]	沈夏闰, 李若楠, 张昊田. 基于CVAE-LSTM的服务器KPI异常检测[J]. 系统工程与电子技术, 2025, 47(3): 1019-1027.
[14]	郭立民, 黄文青, 陈前, 王佳宾. 轻量化的ML-SNet雷达复合干扰识别算法[J]. 系统工程与电子技术, 2025, 47(2): 418-427.
[15]	刘晓琳, 郭梦娇, 李卓. Dueling DQN优化下的航班延误自适应图卷积循环网络预测方法[J]. 系统工程与电子技术, 2025, 47(2): 568-579.