样本不平衡下的海杂波弱目标分类研究  

The Unbalanced Classification of Weak Target in Sea Clutter

作　　者：时艳玲[1] 刘子鹏 贾邦玲 SHI Yanling;LIU Zipeng;JIA Bangling(College of Telecommunications&Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing,Jiangsu 210003,China)

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003

出　　处：《信号处理》2021年第9期1781-1789,共9页Journal of Signal Processing

基　　金：江苏省研究生科研与实践创新计划项目(SJCX19_0249)。

摘　　要：现有的海面弱目标分类算法难以应对单域特征造成特征混叠问题,且存在海杂波和目标样本不平衡的问题。因此,本文研究了一种样本不平衡下的海杂波弱目标分类的方法。首先,从多域提取特征,其中包括从极化域提取球体、双平面和螺旋散射的相对功率特征,从时域提取相对平均幅度特征、和从频域提取非广延熵特征。然后对比分析了海杂波和目标的多域特征之间的区别。由于海杂波特征的样本数目远大于目标样本数目,且海杂波特征具有局部聚集性,为了解决这种样本不平衡以及特征混叠所导致的分类偏差问题,本文设计了一种K均值和支持向量机(SVM)结合的分类器。该分类器主要通过将海杂波样本进行K均值动态聚类,将原本属于一类的海杂波样本分成多类,缓解样本非平衡现象,然后再将多类海杂波样本与目标样本进行SVM分类。经过实测数据验证,该方法具有良好的分类性能。The existing classification algorithms for weak targets at sea clutter are difficult to solve the aliasing of single domain features and the imbalance of samples between sea clutter and targets.In this paper,a classifier for weak targets of sea clutter under sample imbalance is studied.First,the features are extracted from multiple domains,including the relative power features of spheres,biplanes and helical scattering from the polarization domain,the relative mean amplitude features from the time domain,and the non-extensive entropy features from the frequency domain.Then the multi-domain features between sea clutter and target are compared and analyzed.Since the number of samples of sea clutter’s features is much larger than that of target’features,and the sea clutter’s features are local aggregation.In order to solve the classification bias caused by sample imbalance and feature aliasing,a classifier combining K-means and support vector machine(SVM)is designed in this paper.This classifier mainly conducts K-means dynamic clustering of sea clutter samples,divides sea clutter samples originally belonging to one class into multiple classes to alleviate the sample imbalance phenomenon,and then classifies multi-class sea clutter samples and target samples by SVM.The experimental data show that this method has good classification performance.

关 键 词：目标分类 K均值聚类 支持向量机 极化分解 

分 类 号：TN957.52[电子电信—信号与信息处理]