基于极化对比增强和模板匹配的全极化SAR目标分类方法  

Full-polarization SAR Target Classification Method Based on Polarization Contrast Enhancement and Template Matching

作　　者：徐文静 刘杰[2] 于君明 冯晓峰 范睿嘉 尹良 XU Wenjing;LIU Jie;YU Junming;FENG Xiaofeng;FAN Ruijia;YIN Liang(School of Information and Communication Engineering,Beijing University of Posts and Telecommunications,Beijing 100876,China;The 27th Research Instiute of CETC,Zhengzhou 450047,China)

机构地区：[1]北京邮电大学信息与通信工程学院,北京100876 [2]中国电子科技集团公司第二十七研究所,河南郑州450047

出　　处：《无线电工程》2025年第1期138-145,共8页Radio Engineering

摘　　要：随着全极化合成孔径雷达(Synthetic Aperture Radar,SAR)技术的发展,基于机器学习进行全极化SAR目标分类的方法得到广泛应用。小样本全极化SAR数据难以满足深度学习模型对数据量的高要求,会影响目标分类效果。针对这一挑战,借助FEKO电磁仿真软件生成6种飞机模型在3组不同场景下的全极化SAR小样本数据,提出了一种基于极化对比增强和模板匹配的全极化SAR目标分类方法。使用基于拉格朗日乘子法进行极化对比增强的方法得到相对最优极化功率比,构造模板库,在此基础上,提出了一种基于最小差异度原则进行模板匹配的类别判定方法,完成目标分类。解决了原有算法需要设定初始变量、判定不合理等问题。实验结果表明,与原有算法相比目标分类准确率提升了38.87%。With the development of full-polarization Synthetic Aperture Radar(SAR)technology,methods of target classification for full-polarization based on machine learning have been widely applied.However,it is difficult for small-sample full-polarization SAR data to meet high data requirements of deep learning model,which will influence the effect of target classification.To address this challenge,small-sample full-polarization SAR data is generated by using FEKO electromagnetic simulation software,in which six aircraft models are set in three different scenarios,and a full-polarization SAR target classification method based on polarization contrast enhancement and template matching is proposed.Specifically,a polarization contrast enhancement method based on the Lagrange multiplier is used to obtain relative optimum polarization power ratio,and the template library is constructed.On this basis,a category determination method based on the principle of minimum difference degree is proposed to complete target classification.The proposed method addresses the need to set initial variables and the problem of making unreasonable judgments in original algorithm.Experimental results show that target classification accuracy has been improved by 38.87%compared with original algorithm.

关 键 词：极化合成孔径雷达 极化对比增强 模板匹配 目标分类 

分 类 号：TN957.52[电子电信—信号与信息处理]