| 引用本文: | 陈维高,朱卫纲,唐晓婧,贾鑫.栈式降噪自编码器在波形单元识别中的应用[J].哈尔滨工业大学学报,2018,50(11):94.DOI:10.11918/j.issn.0367-6234.201706021 |
| CHEN Weigao,ZHU Weigang,TANG Xiaojing,JIA Xin.The application ofstacked denoising autoencoders in waveform unit identification[J].Journal of Harbin Institute of Technology,2018,50(11):94.DOI:10.11918/j.issn.0367-6234.201706021 |
|
| 摘要: |
| 为提高多功能雷达(Multi-Function Radar, MFR)波形单元的识别准确率和鲁棒性，提出一种栈式降噪自编码器(Stacked Denoising Autoencoders, SDAE)与支持向量机(Support Vector Machine, SVM)相结合的波形单元识别方法.首先摒弃传统依赖脉冲序列分析技术对MFR信号进行处理的方法，通过分析波形单元结构并借助参数间的联合变化特征，提出一种MFR波形单元分段识别模型，将传统对脉冲序列的识别转化为对MFR波形单元的识别；然后在该模型的基础上引入SDAE，对训练样本数据、SDAE隐含层神经元节点进行加噪处理，并利用这些加噪后的样本数据训练优化SDAE网络模型，提取出样本数据的深层稳健特征；最后引入SVM算法，借助SDAE挖掘出的样本深层特征，实现SVM模型的优化，得到最终的波形单元识别模型(SDAE-SVM).仿真实验表明：提出的波形单元识别方法在相同样本数目和测试误差的条件下，与SVM算法相比，能够取得较高的识别准确率，具备更优越的识别效果.证实MFR波形单元识别模型是有效的，且通过SDAE网络的引入，使得SDAE-SVM方法能够自主地挖掘原始信号的深层特征，提高波形单元识别的鲁棒性和准确率.
|
| 关键词: 多功能雷达 波形单元 深度学习 降噪自编码器 支持向量机 |
| DOI：10.11918/j.issn.0367-6234.201706021 |
| 分类号:TN958.92 |
| 文献标识码:A |
| 基金项目: |
|
| The application ofstacked denoising autoencoders in waveform unit identification |
|
CHEN Weigao,ZHU Weigang,TANG Xiaojing,JIA Xin
|
|
(Space Engineering University, Beijing 101416, China)
|
| Abstract: |
| To improve the accuracy and robustness of the Multi-Function Radar (MFR) waveform unit identification, a waveform unit identification method combining Stacked Denoising Autoencoders (SDAE) with Support Vector Machine (SVM) is proposed. The traditional MFR signal processing method relying on pulse sequence analysis is abandoned. A MFR waveform unit segmental identification model is proposed by analyzing the structure of waveform unit and the union variation characteristics of parameters. By using this model, the traditional identification of pulse sequences can be converted to the identification of MFR waveform unit. On the basis of this model, SDAE algorithm is introduced. The training sample data and SDAE hidden layer nodes are processed by noise adding. Using the training sample data, the SDAE network model is trained, and the deep robust feature of the sample data is extracted. Lastly, SVM algorithm is introduced. By using the deep feature of the SDAE output, the SVM model is optimized, and the final waveform unit identification model can be obtained. Simulation results show that under the same condition of sample number and test error, the proposed method can achieve high recognition accuracy, and has better identification results than SVM algorithm. The MFR waveform unit segmental identification model is verified. Besides, through the introduction of SDAE, the proposed SDAE-SVM method can autonomously dig up the deep feature of original data, and improves the robustness and accuracy of the waveform unit identification.
|
| Key words: Multi-Function Radar (MFR) waveform unit Deep Learning Stacked Denoising Autoencoders (SDAE) Support Vector Machine (SVM) |
