基于注意力图剪枝的辐射源信号识别方法

doi:10.12305/j.issn.1001-506X.2025.04.05

系统工程与电子技术 ›› 2025, Vol. 47 ›› Issue (4): 1067-1073.doi: 10.12305/j.issn.1001-506X.2025.04.05

基于注意力图剪枝的辐射源信号识别方法

王慧赋¹^,², 潘海波¹, 罗佳²^,³, 陶诗飞¹^,³^,*

1. 南京理工大学电子工程与光电技术学院, 江苏南京 210094
2. 中国电子科技集团公司第二十九研究所, 四川成都 610036
3. 电磁空间安全全国重点实验室, 四川成都 610036

收稿日期:2024-04-29 出版日期:2025-04-25 发布日期:2025-05-28
通讯作者: 陶诗飞
作者简介:王慧赋 (1998—), 男, 工程师, 硕士, 主要研究方向为辐射源信号识别、深度学习
潘海波 (2000—), 男, 硕士研究生, 主要研究方向为电子侦察、辐射源信号处理
罗佳 (1989—), 男, 高级工程师, 博士, 主要研究方向为电子对抗
陶诗飞 (1987—), 男, 副教授, 博士, 主要研究方向为电子侦察、目标识别、电磁隐身
基金资助:
电磁空间安全全国重点实验室开放基金资助课题

Radiation source signal recognition method based on attention map pruning

Huifu WANG¹^,², Haibo PAN¹, Jia LUO²^,³, Shifei TAO¹^,³^,*

1. School of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
2. The 29th Research Institute of China Electronics Technology Group Corporation, Chengdu 610036, China
3. National Key Laboratory of Electromagnetic Space Security, Chengdu 610036, China

Received:2024-04-29 Online:2025-04-25 Published:2025-05-28
Contact: Shifei TAO

摘要/Abstract

摘要：

针对用于辐射源信号识别的神经网络存在参数冗余、运算量庞大的问题, 提出一种基于注意力图剪枝的辐射源信号识别方法。所提方法利用一阶梯度与特征激活值的乘积来衡量基准网络卷积核的效用值, 并将效用值低的卷积核去除。为避免剪枝后子网络的信号识别性能下降, 所提方法在微调训练阶段引入注意力图知识损失, 并构建联合损失函数, 将基准网络的特征提取能力迁移至子网络。实验结果表明, 对基准网络进行剪枝后, 子网络的信号识别准确率仅下降0.07%, 网络参数量下降48.7%, 运算量下降80.1%，验证了所提方法在保证信号识别准确率的前提下有效地实现了网络轻量化。

关键词: 辐射源信号识别, 模型剪枝, 注意力图, 网络复杂度

Abstract:

In order to solve the problems of redundant parameters and large amount of computation in the neural network used for radiation source signal recognition, a method of radiation source signal recognition based on attention map pruning is proposed. The proposed method uses the product of the first-order gradient and feature activation values to measure the effectiveness values of the benchmark network convolution kernels, and removes the convolution kernels with low effectiveness values. In order to avoid the serious degradation of signal recognition performance of subnetworks after pruning, the proposed method introduces attention map knowledge loss in the fine-tuning training stage and constructs a joint loss function to transfer the feature extraction capability of the benchmark network to the subnetwork. The experimental results show that after pruning the benchmark network, the signal recognition accuracy of the subnetwork decreases by only 0.07%, the number of network parameters decreases by 48.7%, and the amount of computation decreases by 80.1%. These outcomes validate that the approach presented in this paper effectively accomplishes network lightweighting while maintaining the accuracy of signal recognition.

Key words: radiation source signal recognition, model pruning, attention map, network complexity

中图分类号:

TN957.51

王慧赋, 潘海波, 罗佳, 陶诗飞. 基于注意力图剪枝的辐射源信号识别方法[J]. 系统工程与电子技术, 2025, 47(4): 1067-1073.

Huifu WANG, Haibo PAN, Jia LUO, Shifei TAO. Radiation source signal recognition method based on attention map pruning[J]. Systems Engineering and Electronics, 2025, 47(4): 1067-1073.

图/表 10

图1

图2

图3

表1

表2

图4

图5

图6

表3

表4

参考文献 29

1	YANG H H , YANG X P , LIU S K , et al. Radar emitter multi-label recognition based on residual network[J]. Defence Technology, 2022, 18 (3): 410- 417. doi: 10.1016/j.dt.2021.02.005
2	刘赢, 田润澜, 王晓峰. 基于深层卷积神经网络和双谱特征的雷达信号识别方法[J]. 系统工程与电子技术, 2019, 41 (9): 1998- 2005.
	LIU Y , TIAN R L , WANG X F . Radar signal recognition method based on deep convolutional neural network and bispectrum feature[J]. Systems Engineering and Electronics, 2019, 41 (9): 1998- 2005.
3	YAO Y , WANG Z H . Radar signal recognition based on time-frequency and CNN preprocessing[J]. Journal of Detection & Control, 2018, 40 (6): 99- 105.
4	SHI L M , YANG C Z , WANG M L , et al. Recognition method of radar signal modulation method based on deep network[J]. Journal of Ordnance Equipment Engineering, 2021, 42 (6): 190- 193.
5	WANG X B, HUANG G M, ZHOU Z W, et al. Radar emitter recognition based on the short time Fourier transform and convolutional neural networks[C]//Proc. of the 10th International Congress on Image and Signal Processing, 2017.
6	周志文, 黄高明, 高俊, 等. 一种深度学习的雷达辐射源识别算法[J]. 西安电子科技大学学报, 2017, 44 (3): 77- 82.
	ZHOU Z W , HUANG G M , GAO J , et al. Radar emitter identification algorithm based on deep learning[J]. Joumal of Xidian University, 2017, 44 (3): 77- 82.
7	WANG F , YANG C , HUANG S , et al. Automatic modulation classification based on joint feature and convolutional neural network[J]. IET Radar, Sonar & Navigation, 2019, 13 (6): 998- 1003.
8	GAO J P , WANG X , WU R W , et al. A new modulation recognition method based on flying fish swarm algorithm[J]. IEEE Access, 2021, 9, 76689- 76706. doi: 10.1109/ACCESS.2021.3079131
9	LIN A N , MA Z Y , HUANG Z , et al. Unknown radar waveform recognition based on transferred deep learning[J]. IEEE Access, 2020, 8, 184793- 184807. doi: 10.1109/ACCESS.2020.3029192
10	HA P K N, VAN L D, QUANG T D. A parallel neural network-based scheme for radar emitter recognition[C]//Proc. of the 14th International Conference on Ubiquitous Information Management and Communication, 2020.
11	TAO S F , XIAO S , GONG S G , et al. Recognition of electromagnetic signals based on the spiking convolutional neural network[J]. Wireless Communications and Mobile Computing, 2022, 2022, 2395996.
12	LECUN Y , DENKER J , SOLLA S . Optimal brain damage[J]. Advances in Neural Information Processing Systems, 1989, 1, 598- 605.
13	HAN S , POOL J , TRAN J , et al. Learning both weights and connections for efficient neural network[J]. Advances in Neural Information Processing Systems, 2015, 28, 2238772.
14	HAN S , LIU X Y , MAO H Z , et al. EIE: efficient inference engine on compressed deep neural network[J]. ACM SIGARCH Computer Architecture News, 2016, 44 (3): 243- 254. doi: 10.1145/3007787.3001163
15	HAN S, MAO H Z, DALLY W J. Deep compression: compressing deep neural networks with pruning, trained quantization and huffman coding[EB/OL]. [2024-02-22]. https://doi.org/10.48550/arXiv.1510.00149.
16	LIU Z, LI J G, SHEN Z Q, et al. Learning efficient convolutional networks through network slimming[C]//Proc. of the IEEE International Conference Computer Vision, 2017: 2736-2744.
17	LIN S H, JI R H, LI Y C, et al. Accelerating convolutional networks via global & dynamic filter pruning[C]//Proc. of the International Joint Conference on Artificial Intelligence, 2018.
18	PAVLO M, STEPHEN T, TERO K, et al. Pruning convolutional neural networks for resource efficient inference[EB/OL]. [2024-02-22]. https://arxiv.org/abs/1611.06440.
19	巩凯强, 张春梅, 曾光华. 卷积神经网络模型剪枝结合张量分解压缩方法[J]. 计算机应用, 2020, 40 (11): 3146- 3151.
	GONG K Q , ZHANG C M , ZENG G H . Convolution neural network model compression method based on pruning and tensor decomposition[J]. Journal of Computer Applications, 2020, 40 (11): 3146- 3151.
20	王忠锋, 徐志远, 宋纯贺, 等. 基于梯度的深度网络剪枝算法[J]. 计算机应用, 2020, 40 (5): 1253- 1259.
	WANG Z F , XU Z Y , SONG C H , et al. Gradient-based deep network pruning algorithm[J]. Journal of Computer Applications, 2020, 40 (5): 1253- 1259.
21	HE Y H, LIN J, LIU Z, et al. AMC: automl for model compression and acceleration on mobile devices[C]//Proc. of the European Conference on Computer Vision, 2018: 784-800.
22	LIU N, MA X L, XU Z Y, et al. Autocompress: an automatic DNN structured pruning framework for ultra-high compression rates[C]//Proc. of the AAAI Conference on Artificial Intelligence, 2020.
23	LIN M B, JI R, ZHANG Y X, et al. Channel pruning via automatic structure search[EB/OL]. [2024-02-22]. https://doi.org/10.48550/arXiv.2001.08565.
24	李屹, 魏建国, 刘贯伟. 模型剪枝算法综述[J]. 计算机与现代化, 2022 (9): 51- 59.
	LI Y , WEI J G , LIU G W . Survey of model pruning algorithms[J]. Computer and Modernization, 2022 (9): 51- 59.
25	肖帅, 龚帅阁, 李想, 等. FPGA平台轻量化卷积神经网络辐射源信号识别方法[J]. 计算技术与自动化, 2023, 42 (4): 140- 146.
	XIAO S , GONG S G , LI X , et al. Emitter signal identification method with lightweight CNN on FPGA platform[J]. Computing Technology and Automation, 2023, 42 (4): 140- 146.
26	SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[C]//Proc. of the IEEE International Conference on Computer Vision, 2017: 618-626.
27	ZAGORUYKO S, KOMODAKIS N. Paying more attention to attention: improving the performance of convolutional neural networks via attention transfer[EB/OL]. [2024-02-22]. https://doi.org/10.48550/arXiv.1612.03928.
28	LI H, KADAV A, DURDANOVIC I, et al. Pruning filters for efficient convnets[EB/O]. [2024-04-08]. https://arxiv.org/abs/1608.08710.
29	LIN M B, JI R R, WANG Y, et al. Hrank: filter pruning using high-rank feature map[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 1529-1538.

信号类型	参数	取值范围
单频	载频f_c	[1/8, 3/8]f_s
线性调频	起始频率f₀	[1/16, 3/16]f_s
偶二次调频	带宽B	[1/16, 3/16]f_s
余弦调制	起始频率f₀	[1/16, 3/16]f_s
余弦调制	带宽B	[1/20, 1/10]f_s
频率编码	基准频率f₀	[1/40, 1/4]f_s
频率编码	跳频序列N_F	{4, 5, 6}
二相编码	载频f_c	[1/16, 3/16]f_s
二相编码	巴克码长度	{5, 7, 11, 13}
线性调频与频率编码混合调制	基准频率f₀	[1/40, 1/4]f_s
线性调频与频率编码混合调制	子码带宽B	[1/40, 1/20]f_s
线性调频与余弦调制混合调制	起始频率f₀	[1/16, 3/16]f_s
线性调频与余弦调制混合调制	带宽B	[1/16, 3/16]f_s

超参数	取值
训练轮次	40
批次大小	64
优化器	随机梯度下降
学习率	0.01
权重衰减	0.005
动量	0.9

网络类型	参数内存/KB	FLOPs
基准网络	2 179	263.3
子网络	1 392	52.5

信噪比/dB	基准网络	方法1	方法2	方法3	本文所提方法
-8	79.1	71.0	68.9	73.1	78.7
-5	97.6	93.5	92.1	95.2	97.9
-2	99.8	98.6	98.1	99.1	99.7
1	99.9	98.9	98.5	99.5	99.8
4	99.9	99.2	99.1	99.6	99.9

[1]	王慧赋, 梅明飞, 齐亮, 柴恒, 陶诗飞. 基于二值神经网络的辐射源信号识别方法[J]. 系统工程与电子技术, 2024, 46(11): 3912-3919.
[2]	聂千祁, 沙明辉, 朱应申. 基于改进残差神经网络的雷达信号识别方法[J]. 系统工程与电子技术, 2024, 46(10): 3356-3364.
[3]	邵凯, 杜自群, 王光宇. 基于模型剪枝动态调整压缩率的CSI反馈方法[J]. 系统工程与电子技术, 2023, 45(8): 2615-2622.
[4]	徐桂光, 王旭东, 汪飞, 胡国兵, 高涌荇, 罗泽虎. 基于SE-ResNeXt网络的低信噪比LPI雷达辐射源信号识别[J]. 系统工程与电子技术, 2022, 44(12): 3676-3684.
[5]	高诗飏, 董会旭, 田润澜, 张歆东. 基于SRNN+Attention+CNN的雷达辐射源信号识别方法[J]. 系统工程与电子技术, 2021, 43(12): 3502-3509.
[6]	孙艺聪, 田润澜, 王晓峰, 董会旭, 戴普. 基于改进CLDNN的辐射源信号识别[J]. 系统工程与电子技术, 2021, 43(1): 42-47.
[7]	李亚兰, 金炜东, 葛鹏. 基于VMD和特征融合的辐射源信号识别[J]. 系统工程与电子技术, 2020, 42(7): 1499-1503.
[8]	黄颖坤, 金炜东, 余志斌, 吴昀璞. 基于深度学习和集成学习的辐射源信号识别[J]. 系统工程与电子技术, 2018, 40(11): 2420-.
[9]	李序,张葛祥,荣海娜. 基于加权K-近邻法和SVC的雷达辐射源信号识别[J]. Journal of Systems Engineering and Electronics, 2010, 32(6): 1215-1219.

基于注意力图剪枝的辐射源信号识别方法

Radiation source signal recognition method based on attention map pruning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 29

相关文章 9

编辑推荐

Metrics

本文评价