-
摘要:
基于深度学习的雷达干扰感知技术能精确感知各类雷达干扰类型，但需预先构建大规模且完备的训练样本，数据集构建工作量大、难度高，同时存在网络模型参数量较大、计算复杂度高的问题，导致在实际平台中难以应用。针对此问题，提出一种小样本数据驱动的雷达复合干扰轻量化感知网络，结合计算机视觉领域的“目标检测”思想建立干扰感知网络，利用雷达干扰时频分布数据提取多尺度特征图，预置锚框进行回归与分类，使用分组卷积与Ghost卷积对大参数量、高计算量的网络结构进行轻量化改进。实验结果表明：只需小规模的多种单一干扰模式样本，即可实现对单一干扰模式、两两复合模式及3类复合模式的灵活感知，在低干噪比条件下保持较高感知性能的同时大幅压缩模型的参数量与运算量。
Abstract:Radar jamming perception technology based on deep learning can accurately perceive all kinds of radar jamming types, but large-scale and complete training samples need to be constructed in advance. The workload and difficulty of data set construction are large. At the same time, there are some problems such as a large amount of network model parameters and high computational complexity, which make it difficult to apply in the actual platform. This research proposes a lightweight perception network powered by tiny sample data for radar compound jamming in order to overcome this challenge. For the first time, the jamming perception network is established combined with the idea of "target detection" in the field of computer vision. The multi-scale feature map is extracted by using the radar jamming time-frequency distribution data, and the anchor is preset for regression and classification. Secondly, the network structure with large parameters and high computational load is lightweight and improved by using group convolution and ghost convolution. The experimental results show that only a small-scale single jamming mode sample training can realize the flexible perception of single jamming mode, pairwise compound mode and three types of compound mode. The model has a considerably compressed number of parameters and processes while maintaining strong perception performance in the case of a low jamming noise ratio.
-
图 1 4类干扰信号的时频分布图像
Figure 1. Time frequency distribution image of four kinds of jamming signals
图 7 雷达复合干扰灵活感知网络架构
Figure 7. Recognition network architecture of radar compound jamming
图 8 4种网络在不同JNR下的mAP@0.5:0.95
Figure 8. MAP@0.5:0.95 of four networks under different JNR
图 11 两两复合模式干扰感知效果
Figure 11. Jamming perception effect of two kinds of composite modes
图 12 3类复合模式干扰感知效果
Figure 12. Jamming perception effect of three kinds of composite modes
表 1 干扰信号主要仿真参数
Table 1. Jamming signal main simulation parameters
JNR/dB 中心频率/
MHz脉冲重复
周期/μs带宽/MHz 采样频率/
MHz0∶1∶16 10～40 100 40 200 
下载: 导出CSV
表 2 YOLO v5s层级名称及参数量
Table 2. YOLO v5s level names and parameters quantity
层级 名称 参数量 1 Focus 3 520 2 Conv 18 560 3 C3 18 816 4 Conv 73 984 5 C3 156 928 6 Conv 295 424 7 C3 625 152 8 Conv 1 180 672 9 SPP 656 896 10 C3 1 182 720 11 Conv 131 584 12 Upsample 0 13 Concat 0 14 C3 361 984 15 Conv 33 024 16 Upsample 0 17 Concat 0 18 C3 90 880 19 Conv 147 712 20 Concat 0 21 C3 296 448 22 Conv 590 336 23 Concat 0 24 C3 1 182 720 25 Detect 24 273
下载: 导出CSV
表 3 本文网络各模块消融实验的mAP@0.5值
Table 3. MAP@0.5 value of ablation experiments for each module of the proposed network
% Ghost G 参数量/106 浮点运算次数/109 JNR 0 2 4 6 8 10 12 14 16 7.06 16.4 95.5 98.6 98.7 97.8 96.5 98.5 97.6 97.9 99.0 √ 2 5.34 15.4 96.6 97.6 98.0 98.7 97.7 99.4 98.0 98.7 99.4 √ 5.93 15.9 98.3 99.1 99.1 98.9 98.1 99.1 98.3 98.4 99.4 2 6.47 15.9 95.4 99.0 98.9 98.6 98.0 99.3 98.6 98.9 99.5 √ 4 5.05 15.2 97.0 97.8 97.8 97.9 97.4 98.0 97.2 98.6 99.3 4 6.18 15.6 96.6 98.7 98.7 98.5 97.1 99.1 97.3 98.2 99.3
下载: 导出CSV
表 4 本文网络各模块消融实验的mAP@0.5:0.95值
Table 4. MAP@0.5:0.95 value of ablation experiments for each module of the proposed network
% Ghost G 参数量/106 浮点运算次数/109 JNR 0 2 4 6 8 10 12 14 16 7.06 16.4 73.5 75.7 78.9 77.2 76.3 77.7 77.2 76.1 82.3 √ 2 5.34 15.4 74.1 76.6 79.2 78.5 78.1 80.5 77.7 80.6 86.4 √ 5.93 15.9 74.2 75.3 76.8 78.0 76.7 78.8 74.3 77.3 84.8 2 6.47 15.9 73.6 76.1 80.1 79.4 78.7 79.3 78.5 78.8 84.8 √ 4 5.05 15.2 73.6 73.9 77.0 76.7 76.4 78.2 75.0 78.9 84.1 4 6.18 15.6 72.9 76.6 79.0 77.9 76.1 79.1 76.6 77.0 82.6
下载: 导出CSV
表 5 10类复合干扰在不同JNR下的召回率感知结果
Table 5. Recall perceived results of 10 types of composite jamming under different JNR
复合干扰类型 JNR 0 2 4 6 8 10 12 14 16 ISRJ+SMSP 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 CSJ+NAJ 0.86 1.00 1.00 1.00 0.90 1.00 0.90 1.00 1.00 NAJ+SMSP 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 CSJ+ISRJ 0.90 1.00 1.00 1.00 1.00 0.96 1.00 1.00 1.00 CSJ+SMSP 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 ISRJ+NAJ 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 CSJ+ISRJ+SMSP 0.87 0.93 1.00 0.93 1.00 1.00 0.96 1.00 1.00 CSJ+ISRJ+NAJ 0.86 1.00 0.87 0.87 0.92 0.97 0.96 1.00 1.00 ISRJ+NAJ+SMSP 0.96 0.98 1.00 0.95 1.00 1.00 1.00 1.00 1.00 CSJ+SMSP+NAJ 0.97 1.00 0.93 1.00 0.93 1.00 0.98 1.00 1.00
下载: 导出CSV
-
[1] 丁鹭飞, 耿富禄, 陈建春. 雷达原理[M]. 4版. 西安: 西安电子科技大学出版社, 2020:8-13.DING L F, GENG F L, CHEN J C. Radar principle[M]. 4th ed. Xi’an: Xidian University Press, 2020:8-13 (in Chinese). [2] SU D T, GAO M G. Research on jamming recognition technology based on characteristic parameters[C]//2020 IEEE 5th International Conference on Signal and Image Processing. Piscataway: IEEE Press, 2021: 303-307. [3] 熊伟, 曹兰英, 郝志梅. 基于多尺度相像系数的雷达干扰类型频域识别[J]. 计算机仿真, 2010, 27(3): 19-22.XIONG W, CAO L Y, HAO Z M. Frequency recognition of radar jamming types base on multi-scale resemblance coefficient[J]. Computer Simulation, 2010, 27(3): 19-22 (in Chinese). [4] 郝万兵, 马若飞, 洪伟. 基于时频特征提取的雷达有源干扰识别[J]. 火控雷达技术, 2017, 46(4): 11-15.HAO W B, MA R F, HONG W. Radar active jamming identification based on time-frequency characteristic extraction[J]. Fire Control Radar Technology, 2017, 46(4): 11-15 (in Chinese). [5] 李紫航, 宋万杰. 有源干扰感知的熵理论方法[J]. 信号处理, 2017, 33(12): 1652-1656.LI Z H, SONG W J. Entropy theory method for active jamming perception[J]. Journal of Signal Processing, 2017, 33(12): 1652-1656 (in Chinese). [6] 刘国满, 聂旭娜. 一种基于卷积神经网络的雷达干扰识别算法[J]. 北京理工大学学报, 2021, 41(9): 990-998.LIU G M, NIE X N. A radar jamming recognition algorithm based on convolutional neural network[J]. Transactions of Beijing Institute of Technology, 2021, 41(9): 990-998 (in Chinese). [7] QU Q Z, WEI S J, LIU S, et al. JRNet: Jamming recognition networks for radar compound suppression jamming signals[J]. IEEE Transactions on Vehicular Technology, 2020, 69(12): 15035-15045. doi: 10.1109/TVT.2020.3032197 [8] LYU Q Z, QUAN Y H, FENG W, et al. Radar deception jamming recognition based on weighted ensemble CNN with transfer learning[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-11. [9] 张永顺, 童宁宁, 赵国庆. 雷达电子战原理[M]. 2版. 北京: 国防工业出版社, 2010: 95-130.ZHANG Y S, TONG N N, ZHAO G Q. Principle of radar electronic warfare[M]. 2nd ed. Beijing: National Defense Industry Press, 2010: 95-130(in Chinese). [10] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 779-788. [11] HAN K, WANG Y H, TIAN Q, et al. GhostNet: More features from cheap operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 1577-1586. [12] 毕鹏程, 罗健欣, 陈卫卫. 轻量化卷积神经网络技术研究[J]. 计算机工程与应用, 2019, 55(16): 25-35.BI P C, LUO J X, CHEN W W. Research on lightweight convolutional neural network technology[J]. Computer Engineering and Applications, 2019, 55(16): 25-35 (in Chinese). [13] HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 770-778. [14] REDMON J, FARHADI A. YOLOv3: An incremental improvement[EB/OL]. (2018-04-08)[2022-04-10]. https://arxiv.org/abs/1804.02767.pdf. [15] LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot MultiBox detector[C]//European Conference on Computer Vision. Berlin: Springer, 2016: 21-37. -
下载:
点击查看大图
计量
- 文章访问数: 170
- HTML全文浏览量: 23
- PDF下载量: 22
- 被引次数: 0
