摘要
基于半监督学习理论的水下目标识别系统能够从未知类别测试集中识别出已学习类别测试样本,并拒判未学习类别测试样本。描述并讨论了两种基于支持向量数据描述(Support Vector Data Description,SVDD)的半监督水下目标识别算法:半监督SVDD与半监督SVDD集成。利用四类实测水下目标样本进行实验,训练样本为三类已知类别样本,测试样本为四类样本,包含一类未学习类别样本。对两种算法实验结果进行比较,表明半监督SVDD集成算法比半监督SVDD算法能更好地识别已学习类别测试样本,并能有效拒判未学习类别测试样本,不足之处为时间消耗与过程复杂程度比半监督SVDD算法高。
The underwater target recognition system based on the theory of supervised learning can recognize class learned underwater targets’ samples and reject class unlearned underwater targets’ samples from class unknown testing set. This paper describes and discusses two semi-supervised underwater target recognition algorithms based on Support Vector Data Description（SVDD）： Semi-supervised SVDD algorithm（SS-SVDD） and Semi-supervised SVDD ensemble algorithm（SS-SVDDE）. Four different classes of underwater targets’ samples are used in the experiment, the training samples are 3-classes samples, the testing samples are 4-classes samples which contain one class of class unlearned samples. By comparing the two algorithms’ experimental results, it is found that SS-SVDDE algorithm can recognize class learned testing samples better than SS-SVDD algorithm and effectively reject class unlearned testing samples, but SS-SVDDE algorithm’s time consumption and process complexity are higher than SS-SVDD algorithm.
出处
《声学技术》
CSCD
2014年第1期10-13,共4页
Technical Acoustics
关键词
半监督
水下目标识别
类别测试样本
支持向量数据描述
分类器集成
semi-supervised
underwater target recognition
class test samples
Support Vector Data Description（SVDD）
classifier ensemble
