基于解耦表征变分自编码机的雷达目标识别算法  

Radar HRRP Target Recognition Algorithm Based on Variational Auto-encoder with Disentangled Representation

作　　者：张晏合 臧月进 陈渤[1] 徐铭晟 ZHANG Yanhe;ZANG Yuejin;CHEN Bo;XU Mingsheng(National Laboratory of Radar Signal Processing,Xidian University,Xi’an 710071,Shaanxi,China;Shanghai Electro-Mechanical Engineering Institute,Shanghai 201109,China)

机构地区：[1]西安电子科技大学雷达信号处理国家级重点实验室,陕西西安710071 [2]上海机电工程研究所,上海201109

出　　处：《空天防御》2022年第2期87-93,共7页Air & Space Defense

基　　金：国家自然科学基金(61771361,61701379);国家杰出青年科学基金(61525105)。

摘　　要：在雷达高分辨距离像目标识别任务中,数据表征对于分类器的性能影响巨大。针对传统法提取的数据表征可分性不强的问题,本文提出了一种解耦表征变分自编码机的方法。该方法通过对类间特性表征和类间共性表征分别建模的方式,优化网络模型参数,实现了高可分性雷达高分辨距离项数据的表征提取,并采用基于实测数据的实验验证了所提算法的有效性。In radar high-resolution range profile target recognition, the representation of data can influence the performance of the classifier significantly. In order to make the data representation more separable, this paper proposes a variational autoencoder with disentangled representation. By modeling the common representation between different class and the distinctive representation respectively, after optimizing the parameter of the network, it can extract a more separable representation of the data. The measured HRRP data are used to show the effectiveness and efficiency of the algorithm.

关 键 词：雷达自动目标识别 高分辨距离像 特征提取 变分自编码机 

分 类 号：TN958[电子电信—信号与信息处理]