雷达信号工作模式识别研究
发布时间：2018-09-03 09:26
【摘要】：随着科技的迅速发展,电子对抗技术已逐步成为现代军事战争的核心,而雷达工作模式识别又是其重中之重,它识别接收机接收到的敌方雷达信号,并估计其威胁水平,能够为防御和进攻提供准确的情报,是实行电子对抗的前提。雷达信号工作模式识别主要包括复杂雷达信号的分选预处理、单辐射源信号工作模式识别、工作模式威胁等级评估几个部分。随着电磁环境的不断恶化,雷达工作模式识别出现了新的挑战,传统的信号处理技术对新型多功能雷达的识别精度已经逐渐不能满足要求,机器学习能够合理地提取雷达信号特征,准确高效地训练与更新模型,因此,本工作运用机器学习的方法对仿真雷达信号进行工作模式动态识别。本文主要完成以下工作:(1)介绍复杂环境中雷达信号分选的概念和一些传统的分选方法,实现用聚类的方式处理侦查到的复杂雷达信号的分选问题。(2)采用自适应滑动窗口划分方法,对分选得到的单辐射源信号进行划分,相比于传统的固定窗口划分,该方法能够根据模式已知的雷达信号中每种工作模式的长度,自适应地选择出最优的窗口长度与其滑动步长。(3)运用两种分类思路进行窗口序列分类。窗口特征为二维特征,分别是时间维和特征维,第一种思路是在时间维度上进行特征选择,对选择得到的特征向量进行训练。第二种是运用多任务的思想,将雷达模式识别分为多个相互关联的学习任务,多个任务同时并行的学习,最终得到多任务分类器(4)采用一种对雷达序列工作模式的概率进行融合决策的方法,根据相邻窗口序列的分类概率以及概率变化趋势,将前一个窗口的类别因素叠加到当前窗口上,此方法可以实现连续到达雷达信号的动态识别,还可以修正部分误分类窗口序列的类别。(5)将增量学习方法运用到雷达动态识别中,提出雷达的增量学习方法。在雷达种类和工作模式不断增加的情况下,节省了存储空间,加快了模型更新速度。(6)运用C/S模式,多线程的方式实现雷达系统进行前端人机交互、动态显示和后端实时处理。雷达信号越来越复杂,只有准确快速地识别出雷达信号的工作模式,才能快速有效地制定对抗措施。因此研究雷达信号工作模式动态识别技术具有重要的意义。
[Abstract]:With the rapid development of science and technology, electronic warfare technology has gradually become the core of modern military warfare, and radar working pattern recognition is the most important. It recognizes the enemy radar signal received by the receiver and estimates its threat level. The ability to provide accurate information for defense and attack is a prerequisite for electronic countermeasures. The working mode recognition of radar signal mainly includes several parts: sorting and preprocessing of complex radar signal, working mode recognition of single emitter signal and evaluation of threat level of working mode. With the continuous deterioration of electromagnetic environment, new challenges have emerged in radar working pattern recognition. The recognition accuracy of traditional signal processing technology for new multifunctional radar has been gradually unable to meet the requirements. Machine learning can extract radar signal features reasonably, train and update the model accurately and efficiently. Therefore, this work uses machine learning method to dynamically identify the operating mode of simulated radar signal. The main work of this paper is as follows: (1) the concept of radar signal sorting in complex environment and some traditional sorting methods are introduced. The method of clustering is used to deal with the problem of complex radar signal sorting. (2) the method of adaptive sliding window is used to divide the single emitter signal, which is compared with the traditional fixed window partition. This method can adaptively select the optimal window length and its sliding step size according to the length of each operating mode in radar signals with known modes. (3) two kinds of classification methods are used to classify window sequences. The window features are two dimensional features, which are the time dimension and the feature dimension. The first way is to select the features in the time dimension and train the selected feature vectors. The second is to divide radar pattern recognition into several interrelated learning tasks by using the idea of multi-task. Finally, the multitask classifier (4) adopts a fusion decision method for the working mode probability of radar sequence, according to the classification probability and the probability variation trend of the adjacent window sequence. When the category factors of the former window are superimposed on the current window, this method can realize the dynamic recognition of radar signals continuously, and can also modify the category of partial misclassification of window sequences. (5) the incremental learning method is applied to radar dynamic recognition. An incremental learning method for radar is proposed. With the increasing of radar types and working modes, the storage space is saved and the updating speed of the model is accelerated. (6) the radar system realizes the front-end man-machine interaction, dynamic display and back-end real-time processing by using C / S mode and multi-thread mode. Radar signal is becoming more and more complex. Only by identifying the working mode of radar signal accurately and quickly can countermeasures be formulated quickly and effectively. Therefore, it is of great significance to study the dynamic recognition technology of radar signal working mode.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.51
[Abstract]:With the rapid development of science and technology, electronic warfare technology has gradually become the core of modern military warfare, and radar working pattern recognition is the most important. It recognizes the enemy radar signal received by the receiver and estimates its threat level. The ability to provide accurate information for defense and attack is a prerequisite for electronic countermeasures. The working mode recognition of radar signal mainly includes several parts: sorting and preprocessing of complex radar signal, working mode recognition of single emitter signal and evaluation of threat level of working mode. With the continuous deterioration of electromagnetic environment, new challenges have emerged in radar working pattern recognition. The recognition accuracy of traditional signal processing technology for new multifunctional radar has been gradually unable to meet the requirements. Machine learning can extract radar signal features reasonably, train and update the model accurately and efficiently. Therefore, this work uses machine learning method to dynamically identify the operating mode of simulated radar signal. The main work of this paper is as follows: (1) the concept of radar signal sorting in complex environment and some traditional sorting methods are introduced. The method of clustering is used to deal with the problem of complex radar signal sorting. (2) the method of adaptive sliding window is used to divide the single emitter signal, which is compared with the traditional fixed window partition. This method can adaptively select the optimal window length and its sliding step size according to the length of each operating mode in radar signals with known modes. (3) two kinds of classification methods are used to classify window sequences. The window features are two dimensional features, which are the time dimension and the feature dimension. The first way is to select the features in the time dimension and train the selected feature vectors. The second is to divide radar pattern recognition into several interrelated learning tasks by using the idea of multi-task. Finally, the multitask classifier (4) adopts a fusion decision method for the working mode probability of radar sequence, according to the classification probability and the probability variation trend of the adjacent window sequence. When the category factors of the former window are superimposed on the current window, this method can realize the dynamic recognition of radar signals continuously, and can also modify the category of partial misclassification of window sequences. (5) the incremental learning method is applied to radar dynamic recognition. An incremental learning method for radar is proposed. With the increasing of radar types and working modes, the storage space is saved and the updating speed of the model is accelerated. (6) the radar system realizes the front-end man-machine interaction, dynamic display and back-end real-time processing by using C / S mode and multi-thread mode. Radar signal is becoming more and more complex. Only by identifying the working mode of radar signal accurately and quickly can countermeasures be formulated quickly and effectively. Therefore, it is of great significance to study the dynamic recognition technology of radar signal working mode.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.51
【相似文献】
相关期刊论文 前10条
1 ;雷达信号及其处理[J];电子科技文摘;2002年08期
2 ;雷达信号及其处理[J];电子科技文摘;2006年04期
3 李坤;;雷达信号及其监测研究[J];中国无线电;2006年11期
4 张金玉;;传输原因引起的雷达信号故障二例[J];空中交通管理;2009年09期
5 何骏敏;;复杂体制雷达信号的特点分析[J];科技风;2010年16期
6 潘建寿;;，
本文编号：2219566
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2219566.html
