| 摘要: |
| 针对传统目标识别方法人工提取特征难以挖掘到数据深层次特征的问题，提出了将卷积神经网络(CNN)应用于高分辨距离像(HRRP)的目标识别方法，实现了对数据深层次特征的自动提取。首先构造CNN模型，设置网络参数；然后针对HRRP数据是一维的问题，将HRRP数据重新排列使一维数据变为二维数据；其次用训练数据对CNN模型进行训练得到网络参数；最后用训练好的网络模型对测试数据进行目标识别。通过对数据的减半并且添加噪声，验证了CNN的泛化性能。通过对学习率的优化，可以进一步提高CNN的识别率。实测数据的实验结果表明，CNN具有较好的识别性能。 |
| 关键词: 雷达目标识别 高分辨距离像 卷积神经网络 深度学习 |
| DOI：DOI:10.3969/j.issn.1672-2337.2020.01.005 |
| 分类号:TN957.51；TP183 |
| 基金项目:国家自然科学基金（No.61501486，61372165） |
|
| High Resolution Range Profile Ship Target Recognition Based on CNN |
|
ZHANG Qi，LU Jianbin，LIU Tao，LIU Qiyue
|
|
1. School of Electronic Engineering，Naval University of Engineering，Wuhan 430033，China;2. Unit 91245 of PLA，Huludao 125000，China
|
| Abstract: |
| For the problem that traditional target recognition method is difficult to mine the deep feature of data，the application of convolutional neural network (CNN) to high resolution range profile (HRRP) target recognition method is proposed in this paper. It realizes the automatic extraction of deep features of data. Firstly，the CNN model is constructed and the network parameters are set. Secondly，for the problem that HRRP data are one dimensional，HRRP data are reordered to transform one dimensional data into two dimensional data. Then，the CNN model is trained with training data to obtain network parameters. Last，the target recognition of test data is carried out with trained network model. The generalization performance of CNN is verified by halving the data and adding noise. The target recognition rate of CNN can be further improved by optimizing the learning rate. Experimental results of measured data show that CNN achieves better recognition performance. |
| Key words: radar target recognition high resolution range profile(HRRP) convolutional neural network(CNN) deep learning |
