基于注意力机制的SRU模型雷达HRRP目标识别  

Attention-based Simple Recurrent Unit Model for Radar HRRP Target Recognition

作　　者：岳智彬 卢建斌[1] 万露 YUE Zhibin;LU Jianbin;WAN Lu(College of Electronic Engineering,Naval University of Engineering,Wuhan 430033)

机构地区：[1]海军工程大学电子工程学院,武汉430033

出　　处：《舰船电子工程》2023年第4期44-48,共5页Ship Electronic Engineering

摘　　要：针对传统目标识别方法难以提取雷达高分辨距离像(HRRP)的深层特征问题,提出了一种基于注意力机制的简单循环单元(SRU)的模型进行HRRP目标识别。该模型通过SRU快速提取HRRP时序特性,引入自注意力机制自适应对重要特征进行加权,增强隐藏层状态特征的表达能力。同时通过堆叠由SRU,自注意力机制和前馈神经网络组成的模块构建深层网络,对HRRP深层特征自动提取。实验结果表明,对比其他模型,该模型可以有效识别目标。在二维可视化下,提取的深层特征可分性最好。In order to solve the problem that traditional target recognition method is difficult to extract the deep feature of radar High-Resolution Range Profile(HRRP),a model of Simple Recurrent Unit(SRU)based on attention mechanism is proposed for HRRP target recognition.The model quickly extracts HRRP temporal feature by SRU,introduces a self-attention mechanism to adaptively weight important feature,and enhances the expression of hidden layer state feature.Meanwhile the model automatically extracts HRRP deep feature by stacking modules consisting of SRU,self-attentive mechanism and feedforward neural network.The experimental results show that the model can effectively recognize target.Under two-dimensional visualization,the model extracts the best separability of deep feature.

关 键 词：雷达自动目标识别 高分辨距离像 循环神经网络 注意力机制 

分 类 号：TN95[电子电信—信号与信息处理]