| 摘要: |
| 针对雷达真实目标、地杂波和密集假目标的辨识问题，提出了一种基于分解卷积神经网络的雷达目标辨识算法。以深度可分离卷积为基础建立分解卷积神经网络模型。为了减少模型参数，通过减少卷积核数量和全连接层连接节点数量，减少识别特征种类，建立了精简分解卷积神经网络。实测数据的处理结果表明，该算法与现有卷积神经网络方法相比，精简分解卷积神经网络对真实目标样本、地杂波样本和密集假目标样本具有更高的识别正确率，且精简模型参数数量不到现有方法的十分之一。 |
| 关键词: 雷达抗干扰 密集假目标 目标辨识 分解卷积神经网络 |
| DOI：DOI: 10.3969/j.issn.1672-2337.2019.01.016 |
| 分类号:TN974 |
| 基金项目: |
|
| A Factorized Convolutional Neural Network Based Algorithm for Radar Target Discrimination |
|
LUO Heng， LI Zenghui，LI Jianxun，LI Xiaobo
|
|
1.School of Electronic Countermeasures, National University of Defense Technology,Hefei 230031, China;2.2.Strategic Early Warning Research Institute, Air Force Academy, Beijing 100089, China
|
| Abstract: |
| In order to discriminate real targets, clutter, and dense multi-false targets, a factorized convolutional neural network based algorithm for radar target discrimination is proposed. The factorized convolutional neural network model is established based on depthwise separable convolution. To reduce the parameters of the model, the simplified factorized convolutional neural network is set up by reducing the number of convolutional kernel, the connection nodes of fully connected layer, and the categories of discrimination features. The result of the measured data demonstrates that the simplified factorized convolutional neural network has higher discrimination rate for real targets, clutter, and dense multi-false targets compared with the existing model. Its parameters are less than 1/10 of the existing model. |
| Key words: radar anti-jamming dense multi-false targets target discrimination factorized convolutional neural network |
