基于深度置信网络的高分辨率雷达距离像识别  

High resolution range profile recognition based on deep belief network

作　　者：王泓霖 李伟[2] 许强 徐建业 张军[3] Wang Honglin;Li Wei;Xu Qiang;Xu Jianye;Zhang Jun(College of Graduate,Air Force Engineering University,Xi’an 710077,China;College of Information&Navigation,Air Force Engineering University,Xi’an 710077,China;College of Electronic Science&Engineering,National University of Defense Technology,Changsha 410073,China)

机构地区：[1]空军工程大学研究生院,西安710077 [2]空军工程大学信息与导航学院,西安710077 [3]国防科技大学电子科学与工程学院,长沙410073

出　　处：《计算机应用研究》2020年第3期872-875,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目;航天科技创新基金资助项目;航空科学基金资助项目。

摘　　要：为提高雷达目标识别准确率,提出了一种基于深度置信网络(DBN)的高分辨率雷达距离像(HRRP)识别方法。首先利用受限玻尔兹曼机(RBM)对HRRP数据进行逐层无监督训练,根据对比散度(CD)算法更新网络参数,通过误差重构设计DBN深度;而后利用反向传播(BP)机制对DBN模型参数进行有监督的微调;最后基于该模型实现了HRRP的分类与识别。实验结果表明,与传统神经网络相比,本文方法的识别准确率及噪声鲁棒性显著提高,识别准确率可提高8.5%。In order to improve the accuracy of radar target recognition,this paper proposed a high resolution range profile(HRRP)recognition method based on deep belief network(DBN).Firstly,the method used a restricted Boltzmann machine(RBM)to perform layer-by-layer unsupervised training on HRRP data.It updated the network parameters according to the contrast divergence(CD)algorithm,and designed the DBN depth by the reconstruction error method.Then,it fine-tuned the DBN model parameters with supervision through a back propagation(BP)mechanism.Finally,based on the model,it realized the classification and recognition of HRRP.The experimental results show that compared with the traditional neural network,the recognition accuracy and noise robustness of the proposed method are significantly improved,and the recognition accuracy can be improved by 8.5%.

关 键 词：深度置信网络 高分辨距离像 重构误差 目标识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]