基于自注意力的双波段预警雷达微动融合识别

doi:10.12305/j.issn.1001-506X.2023.03.11

摘要/Abstract

摘要：

针对预警雷达对气动目标协同识别的需求, 提出一种自适应权重双输入自注意力残差融合识别方法。通过分析不同波段雷达对气动目标的微动差异性, 在传统卷积块注意力模块(convolutional block attention module, CBAM)残差网络的基础上进行针对性改进, 设计加权双输入CBAM(weighted double input-CBAM, WDI-CBAM)残差结构, 对两种波段的时频图浅层特征自动分配权重并融合, 从而均衡不同波段对目标识别的贡献度。仿真和实测数据处理结果表明, WDI-CBAM残差网络训练代价小, 在信噪比较低及驻留时间较短的情况下识别率高。可视化结果进一步证明了所提方法能够合理分配不同波段输入对气动目标分类的重要性。

关键词: 自注意力机制, 权重自适应, 双波段融合, 气动目标识别

Abstract:

In response to the demand of early warning radar to aerodynamic target cooperative identification, an adaptive weight double-input self-attention residual fusion recognition method is proposed. By analyzing the fretting differences of different band radars on aerodynamic targets, the fusion neural network is modified on the basis of convolutional block attention module (CBAM) residual network. Weighted double input-CBAM (WDI-CBAM) residual structure is designed to automatically assign weight and merge the shallow features of time-frequency maps of two frequencies, so as to balance the contribution of different bands to target recognition. Simulation and measured data processing results show that the WDI-CBAM residual network has a lower training cost and a higher recognition rate under the condition of low signal to noise ratio and short resident time. The visualization results further prove that the reasonably importance of different band inputs is allocated for aerodynamic target classification by the proposed method.

Key words: self-attention mechanism, adaptive weight, double band fusion, aerodynamic target recognition

中图分类号:

TN95

赵庆媛, 赵志强, 叶春茂, 鲁耀兵. 基于自注意力的双波段预警雷达微动融合识别[J]. 系统工程与电子技术, 2023, 45(3): 708-716.

Qingyuan ZHAO, Zhiqiang ZHAO, Chunmao YE, Yaobing LU. Micro-motion fusion recognition of double band early warning radar based on self-attention mechanism[J]. Systems Engineering and Electronics, 2023, 45(3): 708-716.

图/表 13

图1

图2

表1

表2

表3

图3

图4

表4

图5

表5

图6

表6

图7

参考文献 35

15	ZHU J P , CHEN H Q . A hybrid CNN-LSTM network for the classification of human activities based on micro-Doppler radar[J]. IEEE Access, 2020, 8, 24713- 24720. doi: 10.1109/ACCESS.2020.2971064
16	WANG M Y , GUI G L , YANG X B , et al. Human body and limb motion recognition via stacked gated recurrent units network[J]. IET Radar, Sonar & Navigation, 2018, 12 (9): 1046- 1051.
17	LANG Y , WANG Q , YANG Y , et al. Person identification with limited training data using radar micro-Doppler signatures[J]. Microwave and Optical Technology Letters, 2020, 62 (3): 1060- 1068. doi: 10.1002/mop.32125
18	SEYFIOGLU M , GURBUZ S . Deep neural network initialization methods for micro-Doppler classification with low training sample support[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14 (12): 2462- 2466. doi: 10.1109/LGRS.2017.2771405
19	高诗飏, 董会旭, 田润澜, 等. 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. doi: 10.12305/j.issn.1001-506X.2021.12.11
	GAO S Y , DONG H X , TIAN R L , et al. Radar emitter signal recognition method based on SRNN+Attention+CNN[J]. Systems Engineering and Electronics, 2021, 43 (12): 3502- 3509. doi: 10.12305/j.issn.1001-506X.2021.12.11
20	李广帅, 苏娟, 李义红, 等. 基于卷积神经网络与注意力机制的SAR飞机检测[J]. 系统工程与电子技术, 2021, 43 (11): 3202- 3210. doi: 10.12305/j.issn.1001-506X.2021.11.20
	LI G S , SU J , LI Y H , et al. Aircraft detection in SAR images based on convolutional neural network and attention mechanism[J]. Systems Engineering and Electronics, 2021, 43 (11): 3202- 3210. doi: 10.12305/j.issn.1001-506X.2021.11.20
21	曲凌志, 杨俊安, 刘辉, 等. 嵌入注意力机制的通信辐射源个体识别方法[J]. 系统工程与电子技术, 2022, 44 (1): 20- 27.
	QU L Z , YANG J A , LIU H , et al. Method for individual identification of communication radiation source embedded in attention mechanism[J]. Systems Engineering and Electronics, 2022, 44 (1): 20- 27.
22	LI J , JIN K , ZHOU D L , et al. Attention mechanism-based CNN for facial expression recognition[J]. Neurocomputing, 2020, 411, 340- 350. doi: 10.1016/j.neucom.2020.06.014
23	XU H F , CHAI L , LUO Z M , et al. Stock movement predictive network via incorporative attention mechanisms based on tweet and historical prices[J]. Neurocomputing, 2020, 418, 326- 339. doi: 10.1016/j.neucom.2020.07.108
1	CHEN V C , LI F , HO S S , et al. Micro-Doppler effect in radar: phenomenon model and simulation study[J]. IEEE Trans.on Aerospace and Electronic Systems, 2006, 42 (1): 2- 21.
2	FIORANELLI F , RITCHIE M , GRIFFITHS H . Classification of unarmed/armed personnel using the NetRAD multistatic radar for microDoppler and singular value decomposition features[J]. IEEE Geoscience and Remote Sensing Letters, 2015, 12 (9): 1933- 1937. doi: 10.1109/LGRS.2015.2439393
3	RITCHIE M , FIORANELLI F . Multistatic micro-Doppler radar feature extraction for classification of unloaded/loaded micro-drones[J]. IET Radar, Sonar & Navigation, 2017, 11 (1): 116- 124.
4	FIORANELLI F , RITCHIE M , GRIFFITHS H . Centroid features for classification of armed/unarmed multiple personnel using multistatic human micro-Doppler[J]. IET Radar, Sonar & Navigation, 2016, 10 (9): 1702- 1710.
5	ZHANG P F , GANG L , HUO C Y , et al. Classification of drones based on Micro-Doppler radar signatures using dual radar sensors[J]. Journal of Radars, 2018, 7 (5): 557- 564.
6	ZHOU B Y , LIN Y E , KERNEC J L , et al. Simulation framework for activity recognition and benchmarking in different radar geometries[J]. IET Radar, Sonar & Navigation, 2021, 15 (4): 390- 401.
7	CHEN Z X , LI G , FIORANELLI F , et al. Personnel recognition and gait classification based on multistatic micro-Doppler signatures using deep convolutional neural networks[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15 (5): 669- 673. doi: 10.1109/LGRS.2018.2806940
8	ZHANG R , LI G , CLEMENTE C , et al. Multi-aspect micro-Doppler signatures for attitude-independent L/N quotient estimation and its application to helicopter classification[J]. IET Radar, Sonar & Navigation, 2017, 11 (4): 701- 708.
9	CHEN Z X, LI G, FIORANELLI F, et al. Dynamic hand gesture classification based on multistatic radar Micro-Doppler signatures using convolutional neural network[C]//Proc. of the IEEE Radar Conference, 2019.
10	李明, 吴娇娇, 左磊, 等. 基于实测数据的空中目标分类识别算法[J]. 电子与信息学报, 2018, 40 (11): 2606- 2613.
	LI M , WU J J , ZUO L , et al. Aircraft target classification and re-cognition algorithm based on measured data[J]. Journal of Electronics & Information Technology, 2018, 40 (11): 2606- 2613.
11	LI X Y , HE Y , JING X J . A survey of deep learning-based human activity recognition in radar[J]. Remote Sensing, 2019, 11 (9): 1068. doi: 10.3390/rs11091068
24	LEI Y T , DU W W , HU Q H . Face sketch-to-photo transformation with multi-scale self-attention GAN[J]. Neurocomputing, 2020, 396, 13- 23.
25	GAO L L , WANG X H , SONG J K , et al. Fused GRU with semantic-temporal attention for video captioning[J]. Neurocomputing, 2020, 395, 222- 228.
26	WANG Q L, WU B G, ZHU P F, et al. ECA-net: efficient channel attention for deep convolutional neural networks[C]//Proc. of the Conference on Computer Vision and Pattern Recognition, 2020: 11534-11542.
27	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proc. of the European Conference on Computer Vision, 2018: 3-19.
28	PERRY R P, DIPIETRO R C, FANTE R L. Coherent integration with range migration using keystone formatting[C]//Proc. of the IEEE Radar Conference, 2007: 863-868.
29	FU W H , HU Z , LI D . A sorting algorithm for multiple frequency hopping signals in complex electromagnetic environments[J]. Circuits, Systems, and Signal Processing, 2020, 39 (1): 245- 267.
30	陈行勇, 黎湘, 郭桂蓉, 等. 基于旋翼微动雷达特征的空中目标识别[J]. 系统工程与电子技术, 2006, 28 (3): 372- 375.
	CHEN H Y , LI X , GUO G R , et al. Identification of airtarget based on the micromotion radar signatures of blades[J]. Systems Engineering and Electronics, 2006, 28 (3): 372- 375.
31	HU J, SHEN L, SUN G, et al. Squeeze-and-excitation networks[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018.
32	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.
33	ZADEH L A . Review of a mathematical theory of evidence[J]. AI Magazine, 1984, 5 (3): 81- 83.
34	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.
12	GURBUZ S Z , AMIN M G . Radar-based human-motion recognition with deep learning: promising applications for indoor monitoring[J]. IEEE Signal Processing Magazine, 2019, 36 (4): 16- 28. doi: 10.1109/MSP.2018.2890128
13	KIM Y , MOON T Y . Human detection and activity classification based on micro-Doppler signatures using deep convolutional neural networks[J]. IEEE Geoscience and Remote Sensing Letters, 2016, 13 (1): 8- 12. doi: 10.1109/LGRS.2015.2491329
14	KIM B K , KANG H S , PARK S O . Drone classification using convolutional neural networks with merged Doppler images[J]. IEEE Geoscience & Remote Sensing Letters, 2016, 14 (1): 38- 42.
35	TSAO J , STEINBERG B D . Reduction of sidelobe and speckle artifacts in microwave imaging: the CLEAN technique[J]. IEEE Trans.on Antennas and Propagation, 1988, 36 (4): 543- 556.

网络结构	输出尺寸	CBAM+ResNet20	DI-CBAM+ResNet20	WDI-CBAM+ResNet20
主干网络	32×32	[3×3@16]	[3×3@32]	[3×3@16]
卷积块1	32×32	[3×3@16, 2]×3	[3×3@32, 2]×3	[3×3@32, 2]×3
卷积块2	16×16	[3×3@32, 2]×3	[3×3@64, 2]×3	[3×3@64, 2]×3
卷积块3	8×8	[3×3@64, 2]×3	[3×3@128, 2]×3	[3×3@128, 2]×3
全连接层	平均池化, SOFTMAX

目标	w_r/rps	L₁/m	L₂/m	叶片数
1	6	0.69	5.965	4
2	4	1.20	10.965	5
3	6	0	5.345	3
4	20	0	1.900	4
5	23	0	1.800	4
6	39	0.23	0.660	3
7	58	0.38	0.600	38

算法	网络结构	7类目标准确率/%	3大类目标准确率/%	训练轮数	参数量(×10⁶)
1	雷达1+ResNet20	75.5	91.5	50	0.27
2	雷达2+ResNet20	64.2	80.9	55	0.27
3	两部雷达+ResNet20决策级融合	80.1	93.1	—	—
4	雷达1+SE+ResNet20	79.2	92.9	20	0.28
5	雷达2+SE+ResNet20	68.1	83.9	22	0.28
6	两部雷达SE+ResNet20决策级融合	82.9	94.2	—	—
7	雷达1+CBAM+ResNet20	81.8	93.4	15	0.28
8	雷达2+CBAM+ResNet20	71.0	85.1	15	0.28
9	两部雷达CBAM+ResNet20决策级融合	83.1	96.9	—	—
10	DI+ResNet20	84.6	94.7	60	1.09
11	DI-SE+ResNet20	85.9	95.1	25	1.12
12	DI-CBAM+ResNet20	89.7	98.1	20	1.13
13	WDI-CBAM+ResNet20	92.7	99.4	15	1.11
14	DI-CBAM+ResNet32	90.0	98.1	30	1.96
15	DI-CBAM+ResNet44	90.1	98.2	42	2.74
16	DI-CBAM+ResNet56	90.1	98.2	50	3.53
17	WDI-CBAM+ResNet32	92.9	99.5	25	1.94
18	WDI-CBAM+ResNet44	93.1	99.4	36	2.72
19	WDI-CBAM+ResNet56	93.2	99.5	45	3.51
20	雷达1+调制谱特征	49.4	70.2	—	—
21	雷达2+调制谱特征	34.6	61.4	—	—
22	双雷达+调制谱特征	50.7	75.3	—	—
23	雷达1+时频图特征	60.2	80.5	—	—
24	雷达2+时频图特征	56.7	72.9	—	—
25	双雷达+时频图特征	64.6	82.7	—	—

性能	算法	相参积累时间/ms
性能	算法	20	40	60	80	100
7类目标分类准确率	网络7	48.1	72.7	80.9	81.8	88.3
	网络8	29.7	69.6	70.1	71.0	74.8
	网络12	57.1	87.4	88.9	89.7	90.5
	网络13	64.5	89.7	92.1	92.7	93.9
	算法22	20.7	34.9	42.8	50.7	68.7
	算法25	31.5	63.5	70.6	64.6	72.1
3大类目标分类准确率	网络7	68.9	87.6	92.6	93.4	93.5
	网络8	54.1	78.9	84.7	85.1	85.1
	网络12	74.8	88.7	97.8	98.1	98.2
	网络13	85.7	90.8	98.9	99.4	99.5
	算法22	38.9	56.3	60.7	75.3	80.5
	算法25	45.0	60.1	79.4	82.7	84.3

权重平均值	目标1	目标2	目标3	目标4	目标5	目标6	目标7
前16个通道权重平均值	0.81	0.75	0.74	0.67	0.65	0.64	0.40
后16个通道权重平均值	0.52	0.45	0.43	0.54	0.50	0.46	0.30