| 摘要: |
| 雷达工作模式识别是解释雷达行为和功能的基本任务。现有方法难以在信号灵活、环境复杂的条件下筛除脉冲序列中不同空间和不同通道中的冗余信息。本文在深度残差网络的基础上，增加了空间自注意力模块和通道自注意力模块以适应上述信号特点。模型引入自注意力机制以实现雷达序列不同空间和通道的自适应权值分配，使网络能更有效地关注更具差异性的信息，实现了极端条件下雷达工作模式的高精度识别。同经典深度学习网络AlexNet、LeNet、VGGNet、ResNet以及常规深度卷积网络相比，该模型在0~50%漏脉冲条件下，平均识别率提升了36%，在独立测试集40%漏脉冲比例下模型仍然具备90%以上的识别率，证明了所提网络的优越性和有效性。 |
| 关键词: 多功能雷达 模式识别 自注意力机制 特征提取 深度学习 |
| DOI：DOI:10.3969/j.issn.1672-2337.2024.02.007 |
| 分类号:TN971 |
| 基金项目: |
|
| Radar Working Mode Recognition Based on Multi⁃Scale Attention Mechanism ResNet |
|
ZHUO Yihong， XIONG Jingwei， PAN Jifei， GUO Linqing
|
|
College of Electronic Engineering, National University of Defense Technology, Hefei 230037, China
|
| Abstract: |
| Mode recognition is a basic task to interpret radar behavior and function. Under the condition of flexible signal and complex environment, the existing methods are difficult to screen out the redundant information in different spaces and channels in the pulse sequence. In this paper, based on the deep residual network, a spatial self?attention module and a channel self?attention module are added to adapt to the above signal characteristics. The self?attention mechanism is introduced in the model to realize the adaptive weight allocation of different spaces and channels of radar sequence, so that the network can focus on more diverse information more efficiently. The high precision recognition of radar working mode is realized under extreme conditions. Compared with classical deep learning networks such as AlexNet, LeNet, VGGNet, ResNet and conventional deep convolutional networks, the average recognition rate of this model is improved by 36% under the condition of 0~50% leakage pulses. In the independent test set, the model still has a recognition rate of more than 90% under the 40% leakage pulse. The advantages and effectiveness of the proposed network are proved. |
| Key words: multifunctional radar mode recognition self⁃attention mechanism feature extraction deep learning |
