文献类型：期刊文献

中文题名：基于LGBM和深度神经网络的HRRP目标识别方法

英文题名：HRRP Target Recognition Method Based on LGBM and Deep Neural Network

作者：张红莉[1];李月琴[1];韩磊[2];齐英杰[1];张维[1]

第一作者：张红莉

机构：[1]北京联合大学智慧城市学院,北京100101;[2]北京理工大学机电学院,北京100081

第一机构：北京联合大学智慧城市学院

年份：2022

卷号：44

期号：2

起止页码：97-103

中文期刊名：探测与控制学报

外文期刊名：Journal of Detection & Control

收录：CSTPCD;;北大核心:【北大核心2020】；CSCD:【CSCD_E2021_2022】；

基金：北京市自然科学基金青年项目资助(4194078);北京联合大学研究生科研创新项目资助(YZ2020K001)。

语种：中文

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

外文关键词：high resolution range profile;target recognition;feature extraction;deep neural network;light gradient boosting machine

摘要：针对传统的HRRP目标识别方法识别率低、模型泛化能力不足等问题,提出基于轻量级梯度提升机(LGBM)和深度神经网络的HRRP目标识别方法。该方法采用LGBM特征选择算法对提取的HRRP具有明确物理意义、统计特性和平移不变性的特征分量进行二次特征选择,以减少特征冗余和样本维度,有利于目标识别速度的提升;搭建深度神经网络时,为了有效解决过拟合问题,引入Dropout约束,把获得的HRRP目标最优特征样本数据送入深度神经网络分类器进行训练学习和测试,有效提高了模型的泛化能力。仿真实验验证结果表明,在4类雷达目标的分类实验中,所提出的方法在提高识别率的同时,也有效提升了识别速度。
Aiming at the problems of low recognition rate and insufficient model generalization capability of traditional HRRP target recognition methods,an HRRP target recognition method based on LGBM(light gradient boosting achine)and deep neural network was proposed.The method used the LGBM feature selection algorithm to perform secondary feature selection on the extracted HRRP feature components with clear physical significance,statistical properties and translation invariance,in order to reduce feature redundancy and sample dimensionality,which was conducive to the improvement of target recognition speed.When building the deep neural network,the Dropout constraint was introduced to effectively solve the overfitting problem,and the obtained HRRP target optimal sample data was fed into the deep neural network classifier for training and testing,which effectively improved the generalization ability of the model.The simulation experimental results showed that the proposed method effectively improved the recognition rate while increasing the recognition speed in the classification experiments of four types of radar targets.