基于集成间隔优化的对海雷达目标识别算法  

Target recognition method for maritime surveillance radars based on ensemble margin optimization

作　　者：范学满 胡生亮[1] 贺静波[1] Fan Xueman;Hu Shengliang;He Jingbo(Institute of Electronics Engineering, Naval University of Engineering, Wuhan 430033, Chin)

机构地区：[1]海军工程大学电子工程学院,湖北武汉430033

出　　处：《华中科技大学学报（自然科学版）》2017年第12期73-79,共7页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61401493);国家部委基金资助项目(9140A01010415JB11002)

摘　　要：综合考虑对海雷达目标识别的高实时性和强泛化能力要求,提出一种利用模拟退火算法(SA)进行集成间隔优化的静态选择集成(SSE)算法.该算法首先利用SA基于集成间隔最大化搜索出不同大小的最优基分类器子集,然后利用集成分类精确度从中筛选出最终的集成分类器系统.进而提出一种分类器权值、样本权值的迭代求解算法,并考虑这两类权值以及基分类器的分类置信度,给出了8种集成间隔定义.在自建全极化高分辨率距离像(HRRP)分类数据集和17个UCI数据集上分析了集成间隔定义对集成算法性能的影响,通过对比实验验证了该算法的有效性.In consideration of the high demands on real-time performance and generalization ability of the target recognition for maritime surveillance radars,a novel static selective ensemble（SSE）method based on the ensemble margin optimization was proposed.First,optimal subsets of base classifiers with different size were obtained by maximizing the ensemble margin using simulated annealing（SA）,and then,the classification accuracy was used to screen all the candidates to get the final ensemble system.Besides,an iterative algorithm was proposed to calculate the weights of base classifiers and training samples.And then,on the basis of the two kinds of weights as well as classification confidence,eight definitions of ensemble margin were illustrated.The influences of margin definition on the ensemble performance were analyzed using a self-built high resolution range profile（HRRP）dataset and seventeen UCI databases.Finally,the feasibility of the novel algorithm was verified by the contrast experiment.

关 键 词：对海雷达 目标识别 集成间隔 静态选择集成 模拟退火 

分 类 号：TN957.52[电子电信—信号与信息处理]