文献类型：期刊文献

中文题名：基于改进CNN的HRRP目标识别方法

英文题名：Radar HRRP target recognition based on improved CNN

作者：李月琴[1];张红莉[1];张维[1];米雅洁[1];修丽梅[1]

第一作者：李月琴

机构：[1]北京联合大学智慧城市学院,北京100101

第一机构：北京联合大学智慧城市学院

年份：2022

卷号：43

期号：8

起止页码：265-274

中文期刊名：兵器装备工程学报

外文期刊名：Journal of Ordnance Equipment Engineering

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD2021_2022】；

基金：北京市自然科学基金青年项目(4194078);智慧北京各业务信息系统数据结构特征与数据模型详细分类研究(ZB10202004)。

语种：中文

中文关键词：高分辨距离像;雷达目标识别;卷积神经网络;特征提取;轻量级梯度提升机

外文关键词：high resolution range profile;radar target identification;convolutional neural network;feature extraction;light gradient boosting machine

摘要：针对HRRP目标识别的传统识别方法识别率低、模型泛化能力不足,提出了一种适合HRRP样本数据的改进CNN模型;采用一维CNN对HRRP样本进行深层特征提取和目标识别,在构建CNN时引入BN算法加快了损失函数的收敛速度;设计了LGBM分类器作为CNN的分类层,有效提高HRRP识别率和识别速度,进一步提升了模型的识别性能;通过与改进前CNN和传统识别方法的对比实验,结果表明所提的改进CNN在提高目标识别率的同时也有效提升了识别速度,可为后续进行HRRP目标识别提供参考。
To address the problems of low recognition rate,insufficient generalization ability of the model,and that the design of CNN structure still needs to be improved for the traditional recognition method of HRRP target recognition,an improved CNN model suitable for HRRP sample data was proposed.The method adopts a one-dimensional CNN for deep feature extraction and target recognition of HRRP samples,and introduces the BN algorithm in constructing the CNN to accelerate the convergence speed of the loss function.The LGBM classifier was designed as the classification layer of the CNN,which effectively improves the recognition rate and recognition speed of HRRP and further enhances the recognition effect of the model.Through the comparison experiments with the pre-improved CNN and the traditional recognition method,the results show that the proposed improved CNN improves the target recognition rate while effectively enhancing the recognition speed,which can provide a reference for the subsequent HRRP target recognition.