基于SVM的小样本不均衡HRRP舰船目标分类方法  

Small-sample imbalanced HRRP ship target classification method based on SVM

作　　者：查海刚 齐向阳 范怀涛 ZHA Haigang;QI Xiangyang;FAN Huaitao(Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院空天信息创新研究院,北京100190 [2]中国科学院大学,北京100049

出　　处：《现代电子技术》2024年第15期109-114,共6页Modern Electronics Technique

基　　金：中科院空天院科学与颠覆性技术项目(E2Z216010F)。

摘　　要：针对实测HRRP军船民船分类时出现的小样本不均衡问题,提出一种联合Relief算法和PCA算法的特征提取方法,并引入过采样算法及误差迭代加权方法改进SVM分类器。该分类方法对原始高维HRRP图像进行预处理及特征子空间加权,增强了主要特征的可分性,改进的SVM分类器经过迭代加权后分类效果明显提升。作为比较,针对相同实测HRRP舰船目标数据集,分析了自适应增强SVM分类器的分类效果。实验结果表明:提出的改进核空间的迭代加权Smote-SVM分类方法识别效果更好,对高分辨距离像的姿态敏感性具有较好的适应能力。In view of the small-sample imbalance in the classification of measured HRRP(high resolution range profile)military ships and civilian ships,a feature extraction method combining Relief algorithm and PCA(principal component analysis)algorithm is proposed,and the oversampling algorithm and error iterative weighting method are introduced to improve the SVM classifier.In this classification method,the original high-dimensional HRRP image is subjected to preprocessing and feature subspace weight,which enhances the separability of the main features,and the classification effect of the improved SVM classifier is improved significantly after iterative weighting.As a comparison,the classification effect of the adaptive enhanced SVM classifier is analyzed on the same measured HRRP ship target dataset.Experimental results show that the iteratively weighted Smote-SVM classification method with improved kernel space has better recognition effect and adaptability to the attitude sensitivity of HRRP.

关 键 词：高分辨距离像 舰船目标分类 特征提取 支持向量机 改进SVM分类器 PCA算法 

分 类 号：TN709-34[电子电信—电路与系统] TP751.1[自动化与计算机技术—检测技术与自动化装置]