基于改进测地距离的等度规映射及其在声目标特征提取中的应用  

An Improved Isometric Mapping Algorithm Based on New Geodesic Distance and Its Application in the Feature Extraction of Acoustic Targets

作　　者：刘辉[1] 杨俊安[1] 王一[1] 蔡学良[1] 

机构地区：[1]解放军电子工程学院安徽省电子制约技术重点实验室,安徽合肥230037

出　　处：《兵工学报》2012年第10期1178-1184,共7页Acta Armamentarii

基　　金：国家自然科学基金项目(60872113)

摘　　要：经典等度规映射(ISOMAP)算法由于其固有的拓扑不稳定性,容易受噪声影响出现"短路边",导致其在声目标特征提取的应用中性能不佳。提出了一种基于改进测地距离的ISOMAP算法,将近邻图的构造看作是建立一个电路模型,以电路中各节点电容从初始阶段到一定的状态所需要的时间为测地距离的量度,将保持局部结构的鲁棒性与保持全局几何结构的准确性结合在一起,克服了噪声短路点对算法的影响,提高了算法性能。在SensIT实验数据和外场实际采集数据上的实验结果表明,基于改进测地距离的ISOMAP算法的准确性和鲁棒性都有了较大提高。A drawback of the isometric mapping(ISOMAP) algorithm is topological instability,that is,the algorithm is sensitive to the noise,which makes ISOMAP do a bad performance when used in the feature extraction of acoustic targets in the battlefield.To solve the problem,this paper proposes a new method for more robust approximation of the geodesic distances of ISOMAP using a viewpoint of electric circuits.The charge times are used to construct new distance measures.In this way,we achieve both the stability of local methods and the global approximation property of global methods.Experimental results prove that the new distance measures can achieve significant advancement over the former method,which can improve the system performance effectively.

关 键 词：信息处理技术 目标识别 流形学习 等度规映射算法 拉普拉斯特征映射算法 

分 类 号：TN912.3[电子电信—通信与信息系统]