用于水下目标识别的无监督特征选择算法

发布时间：2018-11-10 22:27

【摘要】：随着社会发展,海洋空间对人类变得愈发重要,对新的水下目标自动识别系统的需求也愈发迫切。在水下目标自动识别系统的构建过程中,提取到的特征含有很多冗余特征、不相关特征和噪声特征,影响系统工作效率,降低了分类识别正确率。为此,本文提出一种新的用于水下目标识别的特征选择算法——基于图学习的无监督特征选择算法(Unsupervised Feature Selection Algorithm Based on Graph Learning,UFSGL)。该算法通过同时进行转换矩阵优化和图学习来优化算法框架,并用正则化方法优化加权图中边的光滑度,最后对转换矩阵进行稀疏化从而进行特征选择。使用UCI数据库的sonar数据集对算法性能进行验证,结果证明,UFSGL算法能够有效减少特征子集中的特征个数,并在一定程度上提高分类识别正确率。
[Abstract]:With the development of society, ocean space becomes more and more important to human beings, and the need for new underwater target recognition system becomes more and more urgent. In the process of constructing automatic underwater target recognition system, the extracted features contain many redundant features, irrelevant features and noise features, which affect the working efficiency of the system and reduce the accuracy of classification and recognition. In this paper, a new feature selection algorithm based on graph learning (Unsupervised Feature Selection Algorithm Based on Graph Learning,UFSGL) is proposed for underwater target recognition. The algorithm optimizes the algorithm framework by simultaneous optimization of the transformation matrix and graph learning, and optimizes the edge smoothness of the weighted graph by regularization method. Finally, the transformation matrix is sparse and the feature selection is carried out. The performance of the algorithm is verified by using the sonar data set of UCI database. The results show that the UFSGL algorithm can effectively reduce the number of features in the feature subsets and improve the accuracy of classification and recognition to a certain extent.
【作者单位】：西北工业大学航海学院;
【基金】：中国船舶系统工程研究院水声对抗国防科技重点实验室基金资助项目
【分类号】：TP301.6

【相似文献】