Research on radar target track recognition based on convolutional neural network
-
摘要: 现代战争中雷达信号日趋复杂，如何快速准确地从种类繁多、数据量庞大的雷达检测数据中，获取目标航迹的类别信息，为战场指挥提供准确有效的信息是当前急需解决的难题。传统基于人的经验认知的雷达目标航迹识别方法已经无法有效应对瞬息万变的战场和海量数据。根据实际雷达数据特点，提出了使用对数的雷达航迹预处理方法，并构建了基于卷积神经网络的深度学习模型，实现了对雷达对抗中的目标航迹的识别与检测。基于模拟生成的雷达目标航迹数据对提出的数据预处理方法和构建的模型进行测试；实验表明，所提出的方法能很好地实现对目标航迹的检测与识别。Abstract: A large number of various radar signals in modern warfare make the electromagnetic environment more and more complex. It is urgent to quickly and accurately obtain the category information of the target track from a large number of radar data, and provide accurate and effective information for the battlefield command. The traditional radar-based target recognition method based on human experience or cognition is unable to effectively cope with the ever-changing battlefield and massive data. Based on the characteristics of actual radar data, this paper proposes a logarithmic preprocessing method and constructs a deep learning model based on convolutional neural network. The deep learning model realizes the recognition and detection of the target track in radar confrontation. The built model is tested based on the radar target track data generated by the simulation. Experiments show that the model can effectively detect and identify the target track.
-
表 1 模拟数据示例
Table 1. Example of simulation data
lot number distance/m orientation/(°) height/m time/ms 643 71648 98.15 1190 73079365 647 226790 112.62 8520 73079749 
下载: 导出CSV
表 2 对数预处理数据示例
Table 2. Example of logarithmic preprocessing data
distance 1/m orientation 1/(°) time 1/ms ⋯ distance 30/m orientation 30/(°) time 30/ms 19.614 6 6.915 4 17.890 6 ⋯ 7.327 1 5.971 6 17.896 9 88.811 6 13.689 2 24.167 7 ⋯ 88.791 4 18.393 4 24.169 2
下载: 导出CSV
表 3 CNN网络超参数和优化器设置
Table 3. Hyperparameter and optimizer settings of CNN
epoch batch_size learn_rate dropout activation optimizer fully connected 100 200 0.000 01 0.5 Softmax Adam 128
下载: 导出CSV
表 4 CNN航迹识别准确率
Table 4. Track recognition accuracy of CNN
AUC train set 1 train set 2 test set 1 0.84 0.82 test set 2 0.79 0.77
下载: 导出CSV
表 5 不同数据预处理方法准确率对比表
Table 5. Comparison of accuracy rates of different data preprocessing methods
sample size data processing method accuracy 9000 origin 0.43 9000 normalization 0.66 9000 min 0.85 9000 logarithm 0.96
下载: 导出CSV
表 6 SVM与CNN对比实验结果
Table 6. Comparison of experimental results between SVM and CNN
train set sample size algorithm accuracy train set 1 1500 SVM 0.680 CNN 0.726 train set 2 5000 SVM 0.809 CNN 0.856 train set 2 12000 SVM 0.849 CNN 0.914
下载: 导出CSV
-
[1] 苑传林. 基于FFT信号处理器的舰船雷达目标检测[J]. 舰船科学技术, 2018(8): 76-78. https://www.cnki.com.cn/Article/CJFDTOTAL-JCKX201808027.htmYuan Chuanlin. Ship radar target detection based on FFT signal processor. Ship Science and Technology, 2018(8): 76-78 https://www.cnki.com.cn/Article/CJFDTOTAL-JCKX201808027.htm [2] 于晓涵, 周伟, 关键. 基于特征融合的雷达视频运动目标检测[J]. 雷达科学与技术, 2015(6): 633-638. doi: 10.3969/j.issn.1672-2337.2015.06.013Yu Xiaohan, Zhou Wei, Guan Jian. Radar video moving target detection based on feature fusion. Radar Science and Technology, 2015(6): 633-638 doi: 10.3969/j.issn.1672-2337.2015.06.013 [3] 胡志宏, 张毅. 浅谈基于粒子滤波的微弱雷达目标检测方法[J]. 信息系统工程, 2017(1): 145-145. https://www.cnki.com.cn/Article/CJFDTOTAL-XXXT201701094.htmHu Zhihong, Zhang Yi. Discussion on weak radar target detection method based on particle filter. Information Systems Engineering, 2017(1): 145-145 https://www.cnki.com.cn/Article/CJFDTOTAL-XXXT201701094.htm [4] 王罗胜斌, 徐振海, 刘兴华, 等. 利用复单脉冲比的平面阵列雷达群目标检测方法[J]. 国防科技大学学报, 2018, 40(3): 76-81. https://www.cnki.com.cn/Article/CJFDTOTAL-GFKJ201803012.htmWangluo Shengbin, Xu Zhenhai, Liu Xinghua, et al. Planar array radar target detection method using complex single pulse ratio. Journal of National University of Defense Technology, 2018, 40(3): 76-81 https://www.cnki.com.cn/Article/CJFDTOTAL-GFKJ201803012.htm [5] 罗菁, 段广青, 齐晓光, 等. 基于稀疏超分辨的机载TS-MIMO雷达慢速运动目标检测方法研究[J/OL]. 电光与控制, 2018. http://kns.cnki.net/KCMS/detail/41.1227.TN.20181218.1620.015.html.Luo Jing, Duan Guangqing, Qi Xiaoguang, et al. Research on slow moving target detection method based on sparse super-resolution airborne TS-MIMO radar. Electro-Optic and Control, 2018 [6] 张晓芳. 基于机器学习的激光雷达目标自动检测方法研究[J]. 激光杂志, 2016, 37(10): 137-141. https://www.cnki.com.cn/Article/CJFDTOTAL-JGZZ201610035.htmZhang Xiaofang. Research on automatic detection method of lidar target based on machine learning. Journal of Laser Science, 2016, 37(10): 137-141 https://www.cnki.com.cn/Article/CJFDTOTAL-JGZZ201610035.htm [7] 蔺美青, 李思施. 无源雷达智能目标识别[J]. 国防科技, 2018, 39(2): 63-69. https://www.cnki.com.cn/Article/CJFDTOTAL-GFCK201802010.htmYan Meiqing, Li Sishi. Passive radar intelligent target recognition. National Defense Science and Technology, 2018, 39(2): 63-69 https://www.cnki.com.cn/Article/CJFDTOTAL-GFCK201802010.htm [8] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[C]//International Conference on Neural Information Processing Systems. 2012: 1097-1105. [9] LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. [10] 刘千, 王堃. 基于BP神经网络的防空目标识别方法[J]. 工业仪表与自动化装置, 2015(2): 94-98. https://www.cnki.com.cn/Article/CJFDTOTAL-GYZD201502025.htmLiu Qian, Wang Kun. Anti-aircraft target recognition method based on BP neural network. Industrial Instrumentation & Automation, 2015(2): 94-98 https://www.cnki.com.cn/Article/CJFDTOTAL-GYZD201502025.htm -
点击查看大图
图(3) / 表(6)
计量
- 文章访问数: 2010
- HTML全文浏览量: 503
- PDF下载量: 106
- 被引次数: 0
