支持向量聚类的UUV障碍物孤立点惰性检测  

Outlier inertia detection of UUV obstacles based on support vector clustering

作　　者：严浙平[1] 迟冬南[1] 赵智[1] 赵玉飞[1] 

机构地区：[1]哈尔滨工程大学自动化学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2012年第11期1377-1383,共7页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目(51179038);新世纪优秀人才计划资助项目(NCET-10-0053)

摘　　要：针对UUV避碰声呐探测障碍物过程中数据的弱可观问题,提出了基于支持向量聚类的障碍物孤立点惰性检测方法.利用基于支持向量的聚类算法将声呐探测的数据划分为噪声点,低威胁障碍物和威胁障碍物.由于噪声点和低威胁障碍物不会对UUV的航行造成威胁,所以根据其分布的随机性特点将其作为孤立点进行检测.为了避免探测数据的过早判定导致的对障碍物的过度估计及误判,提出了惰性算法来降低由原始数据的弱可观性和声呐的过度敏感性所带来的障碍物误判的概率.通过仿真试验和海试数据验证表明了该方法对障碍物数据中孤立点检测的有效性.An algorithm of outlier inertia detection of obstacles based on support vector clustering was proposed in order to handle the inaccurate data from sonar. Based on the algorithms of original data clustering findings, the study will explore support vector clustering, which is divided into three groups： noise, non- menacing obstacle and menacing obstacle. Noise and the non-menacing obstacle are no threat to underwater unmanned vehicle （UUV） during navigation and are the distributions of stochastic, therefore, classifying it as the outlier. The inertia algo- rithm avoids the premature use of the data measured by sonar and can decrease the probability of misjudgment for the obstacle induced by the incomplete orignal data. The simulation and the validation support the research derived from data from sea trial effectiveness.

关 键 词：UUV 孤立点检测 支持向量聚类 惰性算法 

分 类 号：TP23[自动化与计算机技术—检测技术与自动化装置]