| 摘要: |
| 传统的雷达高分辨距离像（High Resolution Range Profile,HRRP）序列识别方法依赖于人工提取特征，并且在使用现有的经典深度学习方法识别小数据集时存在梯度消失和过拟合问题，导致收敛速度慢，识别率低。针对上述问题，提出了一种基于注意力机制的集成Inception网络模型，通过集成Attention-Inception单分支网络，实现了HRRP序列更深层次特征的提取；通过对模型的损失函数加入L2正则化，缓解小数据集在集成网络中的过拟合问题；利用InceptionⅠ和InceptionⅡ结构提取HRRP序列多尺度特征，并引入注意力机制计算特征序列的分配权重；加入残差结构，减缓了集成网络梯度消失问题。在预处理后的HRRP序列上进行实验结果表明，所提方法的目标识别率达到93.3%，并且与未去除噪声的HRRP序列相比目标识别率提高了14.67%。 |
| 关键词: 高分辨距离像序列 目标识别 神经网络集成 注意力机制 Inception结构 |
| DOI：10.20079/j.issn.1001-893x.230620002 |
|
| 基金项目:国家自然科学基金资助项目（61871203） |
|
| Radar HRRP Sequence Target Recognition Based on Attention-Inception Network Ensembles |
| FANG Mengyao,ZHANG Zhenkai,LI Wanghua |
| (Ocean College,Jiangsu University of Science and Technology,Zhenjiang 212028,China) |
| Abstract: |
| The traditional high resolution range profile(HRRP) sequence recognition method relies on artificial feature extraction,and when using the existing classical deep learning method to recognize small data sets,there are problems of gradient vanishing and overfitting,which leads to the slow convergence speed and low recognition rate.To solve these problems,an integrated Inception network model based on the attention mechanism is proposed,which realizes the extraction of deeper features of HRRP sequence by integrating the Attention-Inception single-branch network.By adding L2 regularization to the loss function of the model,the overfitting problem of small data sets in the integrated network is alleviated.Inception I and Inception II structures are used to extract multi-scale features of HRRP sequence,and the attention mechanism is introduced to calculate the distribution weight of feature sequence.The residual structure is added to slow down the gradient vanishing problem of integrated network.Experimental results on the preprocessed HRRP sequence show that the target recognition rate of the proposed method reaches 93.3% and is 14.67% higher than that of the HRRP sequence without noise removal. |
| Key words: high resolution range profile sequence target recognition neural network ensembles attention mechanism inception structure |
