摘要
针对水声目标数据的特征冗余问题,提出一种新的近邻无监督特征选择算法。首先利用顺序向后特征搜索算法生成原始特征集的子集,然后利用基于代表近邻选取方法的特征评价机制评价特征子集的优越性。使用实测水声目标数据集和声呐数据集进行特征选择和分类实验,在保持支持向量机平均分类正确率几乎不变的情况下,特征数目分别降低了90%和75%。结果表明,该算法选择出的特征子集,在去除冗余特征后有效地提高了后续学习算法的效率。
The problem of feature redundancy in underwater target recognition has been studying by plenty of researchers. In this paper, a new neighbor based unsupervised feature selection algorithm is proposed. Primarily, the subsets of the original feature set extracted from the dataset are produced by using backward feature searching strategy. Subsequently, these feature subsets are evaluated with the assessment mechanism based on the representative neighbors choosing method. Results of classification experiments with actual measured underwater acoustic target dataset and sonar dataset after feature selection show that the accuracies of SVM classifiers remain almost the same when the numbers of features are decreased by 90% and 75%, which indicates that the proposed method improves the efficiency of subsequent learning algorithm with the redundant features removed.
出处
《声学技术》
CSCD
北大核心
2016年第3期204-207,共4页
Technical Acoustics
基金
水声对抗技术重点实验室开放基金
关键词
水声目标识别
无监督
特征选择
代表近邻
underwater acoustic target recognition
unsupervised
feature selection
representative neighbors