基于深度度量学习的有源欺骗干扰快速识别算法  

Fast Recognition Algorithm of Active Deception Jamming Based on Deep Metric Learning

作　　者：温镇铭 王国宏 于洪波 熊振宇 WEN Zhen-ming;WANG Guo-hong;YU Hong-bo;XIONG Zhen-yu(Institute of Information Fusion,Naval Aviation University,Yantai 264001,China)

机构地区：[1]海军航空大学信息融合研究所,山东烟台264001

出　　处：《中国电子科学研究院学报》2024年第4期307-314,339,共9页Journal of China Academy of Electronics and Information Technology

基　　金：山东省自然科学基金面上资助项目(ZR2020MF015;ZR2020QF010)。

摘　　要：干扰的精准识别是实现干扰抑制的关键前提,但在实际有源欺骗干扰的识别过程中,形态相近的单一干扰易混淆、复合干扰识别准确率不高的问题较为突出。为解决这一问题,文中提出基于深度度量学习的有源欺骗干扰快速识别算法。方法以干扰信号的平滑伪Wigner-Ville分布(Smoothed Pseudo-Wigner-Ville Distribution,SPWVD)作为时频特征样本训练深度度量学习网络,并通过哈希算法和“交叉熵损失函数—三元组损失函数—中心损失函数”的联合约束优化图像特征,以增强深度度量学习网络对时频分布中细微差异的甄别能力。仿真实验表明,经训练后的深度度量学习网络可快速、准确识别八种单一干扰和三种复合干扰,平均识别准确率达到99.36%,且在样本数量较少的情况下依然保持良好性能。The accurate recognition of jamming is the key premise of jamming suppression,but in the process of active deception jamming recognition,the single jamming with similar shape is easy to confuse,and the recognition rate of compound jamming is not high.To solve this problem,this paper proposes an active deception jamming recognition algorithm based on deep metric learning.The method trains a deep metric learning network by using a smooth pseudo Wigner-Ville distribution(SPWVD)of jamming signals as time-frequency feature samples,and the image features were optimized by the joint constraint of hash algorithm and“cross-entropy loss function-triplet loss function-center loss function”to enhance the discrimination ability of the deep metric learning network for subtle differences in time-frequency distribution.Simulation results show that the trained deep metric learning network can quickly and accurately identify 8 kinds of single jamming and 3 kinds of compound jamming,with an average recognition accuracy of 99.36%,and still maintain good performance under the condition of a small number of samples.

关 键 词：有源欺骗干扰 干扰识别 深度度量学习 损失函数 时频分布 

分 类 号：TN947[电子电信—信号与信息处理]