摘要
为在保证目标识别准确率基础上进行有效特征降维,文章以目标识别准确率为特征选择准则,提出一种支持向量机递归特征消除(Support Vector Machine Recursive Feature Elimination,SVM-RFE)快速筛选出部分优质特征子集与猫群算法(Cat Swarm Algorithm,CSO)迭代寻优结合的特征选择方法,并将该方法应用于水声目标识别的特征选择。实验数据处理结果表明:相比SVM-RFE和CSO特征选择算法,文中提出的方法在平均特征维数降低8%的基础上,平均目标识别率提高了1.88%,能够实现有效降维的目的。该方法对判断特征是否适合用于特定的目标识别也有一定应用价值。
In order to reduce the dimension of feature effectively on the basis of ensuring the accuracy of target recognition,a feature selection method based on combining support vector machine recursive feature elimination(SVM-RFE)algorithm and cat swarm optimization(CSO)algorithm is proposed in this paper.The method is applied to feature selection of underwater acoustic target recognition.Experimental data processing results show that:compared with SVM-RFE and CSO feature selection algorithms,the average feature dimension of the proposed method is reduced by 8%,and the average target recognition rate is improved by 1.88%.This method also has a certain application value in judging whether the feature is suitable for specific target recognition or not.
作者
郭政
赵梅
胡长青
GUO Zheng;ZHAO Mei;HU Changqing(Shanghai Acoustics Laboratory,Chinese Academy of Sciences,Shanghai 201815,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《声学技术》
CSCD
北大核心
2021年第1期14-20,共7页
Technical Acoustics
基金
水声对抗技术重点实验室开放基金(JCKY2020207CH02)。
关键词
特征选择
水声目标识别
支持向量机
递归特征消除
猫群算法
feature selection
underwater acoustic target recognition
support vector machine(SVM)
recursive feature elimination(RFE)
cat swarm optimization(CSO)